{"id": "2606.18933", "title": "Zero-Shot Active Feature Acquisition via LLM-Elicitation", "abstract": "Active feature acquisition (AFA) sequentially selects which features to observe to reach a classification or ranking decision. Its central limitation is reliance on large amount of labeled data to fit probabilistic models guiding acquisition. Large language models (LLMs) supply unsupervised domain knowledge, but are poor sequential planners. Asking one to both know and decide conflates capabilities best kept separate. Here, we develop a framework for zero-shot AFA through disciplined elicitation: asking the LLM only for what it can be trusted to return, the unary deviations and pairwise co-variations that are the sufficient statistics of a Markov random field (MRF). We apply our framework to two settings: binary classification and top-$k$ identification. In practice, the LLM reliably returns only discriminative statistics, what distinguishes the classes rather than each class in isolation, which precludes classical AFA. We apply a maximum-entropy closure that resolves this gauge ambiguity. We evaluate on a cohort of Inflammatory Bowel Disease (IBD) patients, an active clinical setting where diagnostic ambiguity and patient heterogeneity obstruct stable treatment strategies. Our framework outperforms the LLM both on real labels and on its own extracted beliefs. Where it matters most, on the hardest patients, our top-$k$ acquisition policy markedly outperforms all existing methods.", "date": "2026-06-17", "categories": []}
{"id": "2604.13899", "title": "Do We Still Need Humans in the Loop? Comparing Human and LLM Annotation in Active Learning for Hostility Detection", "abstract": "Instruction-tuned LLMs can annotate thousands of instances at low cost. This raises two questions for active learning (AL): can LLM labels replace human labels within the AL loop, and does AL remain necessary when entire corpora can be cheaply labeled? We investigate both on a new dataset of 277,902 German political TikTok comments (25,974 LLM-labeled, 5,000 human-annotated), comparing LLM and human annotation across seven conditions, four encoders, and 10 random seeds. Under a two-question interface that mirrors the human annotation task, LLM annotation at scale outperforms human-supervised classifiers at roughly one-tenth the cost (\\$28 for GPT-5.2 Batch API vs. \\$316 for Prolific). The advantage holds for both a closed-source (GPT-5.2) and an open-weight (Qwen3.5-122B-10B) LLM, is robust under soft-label evaluation, and is unlocked specifically by the two-question decomposition; a holistic single-prompt baseline only ties with human supervision. AL provides no reliable advantage over random sampling under either LLM annotator. However, error structure varies sharply: only GPT-5.2 under the two-question interface produces classifiers with near-human FP/FN balance, while other LLM variants over-flag border-control and economic competition discourse. We release the dataset and code.", "date": "2026-04-15", "categories": []}
{"id": "2206.10188", "title": "Analysis of Self-Supervised Learning and Dimensionality Reduction Methods in Clustering-Based Active Learning for Speech Emotion Recognition", "abstract": "When domain experts are needed to perform data annotation for complex machine-learning tasks, reducing annotation effort is crucial in order to cut down time and expenses. For cases when there are no annotations available, one approach is to utilize the structure of the feature space for clustering-based active learning (AL) methods. However, these methods are heavily dependent on how the samples are organized in the feature space and what distance metric is used. Unsupervised methods such as contrastive predictive coding (CPC) can potentially be used to learn organized feature spaces, but these methods typically create high-dimensional features which might be challenging for estimating data density. In this paper, we combine CPC and multiple dimensionality reduction methods in search of functioning practices for clustering-based AL. Our experiments for simulating speech emotion recognition system deployment show that both the local and global topology of the feature space can be successfully used for AL, and that CPC can be used to improve clustering-based AL performance over traditional signal features. Additionally, we observe that compressing data dimensionality does not harm AL performance substantially, and that 2-D feature representations achieved similar AL performance as higher-dimensional representations when the number of annotations is not very low.", "date": "2022-06-21", "categories": []}
{"id": "2606.17805", "title": "QueryMarket: Cost-Aware Online Active Learning in Data Markets", "abstract": "Data acquisition is a major bottleneck for learning in real-time streams: analysts must decide on the fly which labels to purchase while respecting a rolling budget. However, existing online active learning rarely unifies pricing, information gain, and rolling budget constraints under concept drift. We introduce QueryMarket, a market-inspired framework that queries each incoming data point based on its estimated utility to the model and its price. Within this framework, we propose OVBAL (online variance-based active learning), which integrates data pricing with information-driven selection by estimating each sample's marginal utility via a D-optimality criterion with exponential forgetting and executing cost-aware purchases under rolling budget constraints. OVBAL yields a simple, fully online decision rule that adapts to nonstationary streams and heterogeneous label costs. Experiments on synthetic data and a real-world solar power generation forecasting task show that OVBAL is particularly effective under seller-centric pricing and yields a more favorable long-run error-cost trade-off in the real-world task under both pricing schemes.", "date": "2026-06-16", "categories": []}
{"id": "2606.16045", "title": "Active Learning with Low-Rank Structure for Data Selection", "abstract": "In the data selection problem, the objective is to choose a small, representative subset of data that can be used to efficiently train a machine learning model. Sener and Savarese [ICLR 2018] showed that, given an embedding representation of the data and suitable geometric assumptions, heuristics based on $k$-center clustering can be used to perform data selection. This perspective was further explored by Axiotis et. al. [ICML 2024], who proposed a data selection approach based on $k$-means clustering and sensitivity sampling. However, these methods rely on the assumption that the dataset exhibits intrinsic geometric structure that can be effectively captured by clustering, whereas many modern datasets instead possess global algebraic structure that is better exploited by low-rank approximation or principal component analysis.\n  In this paper, we introduce a new data selection framework based on low-rank approximation and residual-based sampling, formulated through the lens of row subset selection and loss-preserving coreset construction. Given an embedding representation of the data satisfying mild regularity conditions, which can be interpreted as algebraic or angular notions of Lipschitz continuity, we show that it is possible to select a weighted subset of $\\tilde{O}\\left(k + \\frac{1}{\\varepsilon^2}\\right)$ data points whose average loss approximates the average loss over the full dataset within a $(1+\\varepsilon)$ relative error, up to an additive $\\varepsilon \\Phi_k$ term, where $\\Phi_k$ denotes the optimal rank-$k$ approximation cost of the embedding matrix. We complement these theoretical guarantees with empirical evaluations, demonstrating that on a range of real-world datasets, our data selection approach achieves improved performance over prior strategies based on uniform sampling or clustering-based sensitivity sampling.", "date": "2026-06-14", "categories": []}
{"id": "2510.00375", "title": "Multidimensional Bayesian Active Machine Learning of Working Memory Task Performance", "abstract": "While adaptive experimental design has outgrown one-dimensional, staircase-based adaptations, most cognitive experiments still control a single factor and summarize performance with a scalar. We show a validation of a Bayesian, two-axis, active-classification approach, carried out in an immersive virtual testing environment for a 5-by-5 working-memory reconstruction task. Two variables are controlled: spatial load L (number of occupied tiles) and feature-binding load K (number of distinct colors) of items. Stimulus acquisition is guided by posterior uncertainty of a nonparametric Gaussian Process (GP) probabilistic classifier, which outputs a surface over (L, K) rather than a single threshold or max span value. In a young adult population, we compare GP-driven Adaptive Mode (AM) with a traditional adaptive staircase Classic Mode (CM), which varies L only at K = 3. Parity between the methods is achieved for this cohort, with an intraclass coefficient of 0.755 at K = 3. Additionally, AM reveals individual differences in interactions between spatial load and feature binding. AM estimates converge more quickly than other sampling strategies, demonstrating that only about 30 samples are required for accurate fitting of the full model.", "date": "2025-10-01", "categories": []}
{"id": "2604.15173", "title": "Boundary-Centric Clip-Budgeted Active Learning for Temporal Action Segmentation", "abstract": "Temporal action segmentation (TAS) in untrimmed videos requires dense temporal supervision. However, most of the annotation cost is spent identifying action transitions where segmentation errors concentrate and small temporal shifts can disproportionately degrade segment-level metrics. We introduce B-ACT, a clip-budgeted active learning framework that explicitly allocates supervision to these error-prone boundary regions. B-ACT operates in a hierarchical two-stage loop: (i) it ranks and queries unlabeled videos using predictive uncertainty, and (ii) within each selected video, it detects candidate transitions from the current model predictions and selects the top-$K$ boundaries via a novel boundary score. The boundary score fuses neighborhood uncertainty, class ambiguity, and temporal prediction dynamics to reveal the underlying importance of each frame. Importantly, our annotation protocol requests labels only at the boundary frames while still training on boundary-centered clips to exploit temporal context through the model's receptive field. Extensive experiments on GTEA, 50Salads, and Breakfast demonstrate that boundary-centric supervision delivers strong label efficiency and consistently surpasses representative TAS active learning baselines and prior state of the art under sparse budgets. Gains are largest on datasets where performance is highly sensitive to boundary placement, as measured by edit and overlap-based F1 metrics.", "date": "2026-04-16", "categories": []}
{"id": "2606.13859", "title": "Closed-loop discovery of out-of-distribution processing protocols by evolutionary search and uncertainty-aware learning", "abstract": "Many materials and chemical systems exhibit history-dependent responses, where functional outcomes are governed not only by final-state variables but by the time-dependent sequence of fields, temperatures, or chemical potentials applied during operation. Discovering new processing protocols is therefore a high-dimensional search problem in which the control variable is an entire waveform or sample history, and conventional strategies either remain confined to conservative interpolative families or become prohibitively measurement intensive. Here, a closed-loop workflow is introduced that couples evolutionary search over a compact waveform representation with uncertainty-aware deep kernel learning to generate, rank, and experimentally validate candidate protocols. Applied to ferroelectric thin films, with the scanning-probe tip-bias waveform as the protocol and the nonlinear electromechanical response as the reward, the workflow discovers waveform families that enhance nonlinearity by de-aging the film. Spatially resolved before/after measurements show that the best-performing waveforms selectively activate pre-existing, weakly pinned domain-wall segments, whereas the worst drive long-range irreversible switching. This framework reframes protocol tuning as out-of-distribution discovery, generalizable to synthesis and annealing trajectories, battery formation protocols, and other high-dimensional control problems.", "date": "2026-06-11", "categories": []}
{"id": "2606.14690", "title": "A Complexity Measure for Active Learning in Multi-group Mean Estimation", "abstract": "We study a \\emph{max-risk} objective for active learning in a multi-group mean estimation $d$-armed bandits: a learner adaptively allocates a budget of $T$ samples across $d$ groups to minimize the worst-case uncertainty index $\\max_{k\\in[d]}\\sigma_k^2/n_k$, where $\\sigma_k$ is the standard deviation of the distribution of arm $d$, and $n_k$ is the number of times arm $d$ is sampled. We develop a local minimax framework and prove the first general lower bound for this objective, valid for any finite-variance hypothesis class. The bound separates difficulty into three orthogonal factors: a \\emph{budget} term, a \\emph{heteroscedasticity} index measuring how unevenly the uncertainty is spread across arms, and a model-dependent complexity measure, the \\emph{Variance Local Curvature} ($\\mathrm{VLC}$), which captures how much information a local change of variance creates inside the hypothesis class. For smooth classes, the $\\mathrm{VLC}$ is a reparametrization of a variance--Fisher information, with closed-form values for common families. Benchmarking against the strongest available upper bound shows near-optimality up to logarithmic factors in broad regimes, and pinpoints a systematic gap in highly heterogeneous instances. Our proof introduces two key ingredients: a loss-induced $\\ell_1$ geometry on the decision space, and a representation-based instance generator that reduces hard-instance construction to an explicit random matrix calculation.", "date": "2026-06-12", "categories": []}
{"id": "2606.12182", "title": "How Low Can You Go? Active Learning for Sparse Model Discovery in the Ultra-Low-Data Limit", "abstract": "Identifying the governing equations of complex dynamical systems remains a fundamental challenge across science and engineering. While early approaches relied on empirical data and heuristics, modern data-driven methods offer greater flexibility and fewer assumptions. However, data acquisition in real-world settings is often expensive. This work addresses this challenge by introducing an active learning strategy for dynamics discovery in the ultra-low data limit. Rather than sampling randomly, our method iteratively prioritizes regions that are most informative for model identification. This approach builds on Sparse Identification of Nonlinear Dynamics (SINDy), and utilizes an ensemble extension, E-SINDy, to estimate epistemic uncertainty and guide the sampling for both ordinary and partial differential equations (ODEs/PDEs). For ODEs, an exhaustive analysis is conducted on the Lorenz system across varying data budgets and noise levels. For PDEs, two systems with contrasting dynamical characteristics are examined: the Burgers' equation, where a sharp shock front creates a distinction between informative and uninformative regions, and the Kuramoto-Sivashinsky equation, which presents a more spatially complex sampling landscape. Across all scenarios, the proposed method accurately identifies the governing dynamics with significantly fewer data samples than random sampling.", "date": "2026-06-10", "categories": []}
{"id": "2606.12386", "title": "ATLAS: Active Theory Learning for Automated Science", "abstract": "Advancing scientific understanding through mechanistic modeling requires posing the right experimental questions to yield maximally informative data. To automate this pursuit within cognitive science, we introduce ATLAS (Active Theory Learning for Automated Science), an active learning framework for the data-driven discovery of interpretable behavioral models. ATLAS iterates between generating mechanistic hypotheses--instantiated as a diverse ensemble of sparse neural networks (Disentangled RNNs)--and designing experiments that optimally distinguish between them. We test this approach on the problem of recovering reinforcement learning agents from their behavior in bandit tasks. ATLAS designs varied sequences of qualitatively novel experiments with temporal structure tailored to underlying agent characteristics. The models trained on these experiments are evaluated against a comprehensive set of metrics for mechanistic modeling that capture behavioral, structural, and computational similarity. ATLAS achieves a 5-10x improvement in sample efficiency across all metrics compared to random experimentation, and its performance is further validated against expert-designed experiments derived from literature. These in silico results showcase ATLAS's potential to accelerate human-interpretable insights in cognitive science and other domains where scientific inquiry relies on discovering mechanistic models.", "date": "2026-06-10", "categories": []}
{"id": "2510.08906", "title": "Gradient-Guided Furthest Point Sampling for Robust Training Set Selection", "abstract": "Training set sampling methods are used to improve model performance and lower data costs in machine learning problems relevant to chemistry. We introduce Gradient Guided Furthest Point Sampling (GGFPS), a simple extension of Furthest Point Sampling (FPS) that leverages molecular force norms to guide efficient sampling of configurational spaces of molecules. Numerical evidence is presented for a toy system (the Styblinski-Tang function) as well as for molecular dynamics trajectories from the MD17 dataset. Our numerical results indicate superior data efficiency and model robustness when using GGFPS compared to FPS and uniform random sampling (URS), as well as established supervised FPS-style selectors, PCov-FPS and PCov-CUR. Distribution analysis of the MD17 data suggests that FPS systematically under-samples equilibrium geometries, resulting in large test errors for relaxed structures. GGFPS cures this artifact and (i) enables up to twofold reductions in training cost without sacrificing predictive accuracy compared to FPS in the 2-dimensional Styblinski-Tang system, (ii) systematically lowers prediction errors for equilibrium as well as strained structures in MD17, and (iii) systematically decreases prediction error variances across all of the MD17 configuration spaces. These results suggest that gradient-aware sampling methods hold great promise as effective training set selection tools, and that naive use of FPS may result in imbalanced training and inconsistent prediction outcomes.", "date": "2025-10-10", "categories": []}
{"id": "2606.09949", "title": "Learning Where to Simulate: Generative Active Sampling for Online PDE Surrogate Training", "abstract": "Data-driven PDE surrogates are trained with data produced by numerical PDE solvers. However, when the surrogate's goal is to generalize across a wide range of PDE configurations (e.g., initial conditions and physical coefficients), generating a representative training set is non-trivial. Uniform sampling of configuration parameters often under-represents trajectories exhibiting challenging dynamics, leading to high prediction errors and large error variance in the trained surrogate. Online training, where data generation and surrogate training are coupled, offers a natural advantage by allowing solver parameters to be steered on-the-fly. To efficiently exploit this capability, we introduce Online Generative Active Sampling (OGAS), an active learning method that reactively learns the relationship between configuration parameters and surrogate performance to control the sampling distribution. OGAS trains a fast diffusion model in parallel to the surrogate to act as a conditional sampler, mapping a surrogate-derived difficulty signal (e.g., loss or uncertainty) to configuration parameters. By actively drawing target signals from a prior biased toward high difficulty, OGAS continuously steers data generation toward challenging regimes without delaying the training workflow. We evaluate OGAS across 2D PDEs with distinct challenging dynamics (Kuramoto-Sivashinsky, Navier-Stokes, Gray-Scott) and up to 308 parameters, using multiple surrogate architectures. Across all settings, OGAS consistently improves tail statistics, yielding substantial reductions in errors above the 99th percentile and overall error dispersion compared to uniform sampling. While prioritizing challenging trajectories introduces a trade-off with average error, OGAS effectively ensures worst-case reliability of trained surrogates with negligible wall-time overhead.", "date": "2026-06-08", "categories": []}
{"id": "2606.10125", "title": "Robust Active Learning for Few-Shot Example Selection in Text-to-SQL", "abstract": "Few-shot example retrieval is the dominant paradigm for grounding large language models (LLMs) in domain-specific text-to-SQL systems. However, the quality of the annotated example bank directly governs system accuracy, and expert annotation is prohibitively expensive. We formalize the active selection of these examples as a constrained experimental design problem over the intrinsic, low-dimensional manifold of semantic query embeddings. Unlike standard active learning frameworks, our setting introduces three critical challenges: varying, query-dependent annotation reliability (heteroscedasticity), strict requirements for spatial diversity across semantic topics (partition matroid constraints), and the inherent reality that the true covariance structure of the embedding space is unknown (misspecification). To address these, we propose a stratified greedy algorithm that maximizes a heteroscedastic mutual information objective. We prove that this objective remains submodular and approximately monotonic on the intrinsic manifold, yielding a theoretical constant-factor approximation guarantee. We establish a spectral bound demonstrating that this approximation guarantee degrades gracefully, rather than catastrophically, when the assumed surrogate kernel diverges from the true underlying data-generating process. Empirical results demonstrate that the proposed strategy significantly reduces labeling effort while maintaining high text-to-SQL retrieval accuracy.", "date": "2026-06-08", "categories": []}
{"id": "2606.11130", "title": "Robust Regression of General ReLUs with Queries", "abstract": "We study the task of agnostically learning general (as opposed to homogeneous) ReLUs under the Gaussian distribution with respect to the squared loss. In the passive learning setting, recent work gave a computationally efficient algorithm that uses $poly(d,1/\\epsilon)$ labeled examples and outputs a hypothesis with error $O(opt)+\\epsilon$, where $opt$ is the squared loss of the best fit ReLU. Here we focus on the interactive setting, where the learner has some form of query access to the labels of unlabeled examples. Our main result is the first computationally efficient learner that uses $d polylog(1/\\epsilon)+\\tilde{O}(\\min\\{1/p, 1/\\epsilon\\})$ black-box label queries, where $p$ is the bias of the target function, and achieves error $O(opt)+\\epsilon$. We complement our algorithmic result by showing that its query complexity bound is qualitatively near-optimal, even ignoring computational constraints. Finally, we establish that query access is essentially necessary to improve on the label complexity of passive learning. Specifically, for pool-based active learning, any active learner requires $\\tilde{\\Omega}(d/\\epsilon)$ labels, unless it draws a super-polynomial number of unlabeled examples.", "date": "2026-06-09", "categories": []}
{"id": "2606.07630", "title": "Active Learning with Foundation Model Priors: Efficient Learning under Class Imbalance", "abstract": "Real-world datasets across image and text domains are often characterized by skewed class distributions and noisy annotations, which jointly degrade model performance, particularly on minority classes. Among existing solutions, active learning offers an effective and efficient paradigm by selectively querying the most informative and balanced samples for annotation. We propose an innovative active learning framework that mitigates class imbalance and selects the most informative samples to annotate. Leveraging foundation model priors, our algorithm enables imbalance-aware co-decisions between foundation model and small model to tackle noisy and imbalanced labels across various domains. We introduce the first study to systematically explore active learning under the dual challenges of label noise and class imbalance across image and text domains. Extensive experiments on imbalanced datasets demonstrate that our method achieves substantial annotation savings-over 50% compared to the best active learning baseline-while preserving performance and robustness to label noise.", "date": "2026-05-30", "categories": []}
{"id": "2606.07910", "title": "CAAL: Contextual Bandits based Online Hand-Craft Active Learning Strategy Selection", "abstract": "The challenge with active learning algorithms is the uncertainty of the statistical distribution of unlabeled data, making it difficult to choose the best hand-crafted strategy. To address this, we introduced Contextual Adaptive Active Learning (CAAL). In CAAL, each \"arm\" represents a hand-crafted strategy. Unlike existing frameworks that select strategies based only on feedback from labeled data, we dynamically choose strategies for labeling batches of data using reward prediction with external context information. This general framework allows for customization with domain knowledge to design more effective rewards and context candidates. In addition, we experimentally show that CAAL outperforms the existing baseline adaptive strategy on public datasets using our reward and context design. Our results are consistent regardless of batch size in each iteration.", "date": "2026-06-06", "categories": []}
{"id": "2606.08148", "title": "Inverse design of bespoke interatomic potentials via active learning by information-matching", "abstract": "Interatomic potentials (IPs) enable large-scale atomistic simulations beyond the reach of first-principles methods, but their predictive reliability depends critically on the selection of training data, quantified uncertainty, and model expressiveness. Active learning (AL) provides a principled framework for constructing efficient and accurate IPs, yet most strategies reduce parameter uncertainty without explicitly accounting for the specific material properties being predicted. The information-matching (IM) approach addresses this limitation by requiring that the selected training data provide at least as much parameter space information as needed to achieve prescribed uncertainty targets for selected quantities of interest (QoIs). Here, we apply IM to develop bespoke IPs specifically tailored for predicting plastic strength in metals. Due to the high computational cost of simulating plastic strength, we employ an indirect IM strategy that targets inexpensive intermediate QoIs that correlate with strength. The IM method enables precise parameter constraints with minimal training data, yielding precise predictions for both the intermediate QoIs and plastic strength. Yet, model error remains a key limitation, and a post hoc uncertainty inflation correction provides a viable means to mitigate this limitation. These findings illustrate both the promise and limits of uncertainty-aware AL for predicting complex material properties.", "date": "2026-06-06", "categories": []}
{"id": "2606.08718", "title": "Deep Active Re-Labeling: Toward Noise-Resilient Annotation Efficiency", "abstract": "While Deep Active Learning (DAL) effectively reduces human annotation costs, its efficacy is constrained by human annotation errors. This is because the data sampled for active learning is assumed to be highly informative for training. When human annotators introduce errors into this informative data at a certain rate, the active learning performance drops significantly and, in some cases, even exhibits worse outcomes than passive learning. In this paper, we first analyze the impact of human annotation errors in the DAL setting. Then we propose a framework to address the human annotation noise problem for DAL. Informed by human learning patterns, the core idea of our proposed solution involves allocating a portion of the human annotation budget to re-annotate data that has already been labeled. Previous theoretical work suggests that when the model possesses a certain level of ability to identify potentially noisy data, even re-labeling a small fraction of the data can effectively remove noise from the active training set. To achieve this, we implement two active noise sampling strategies to detect noise under different circumstances and allocate a part of the annotation budget to re-annotate these instances. Our approach imbues active learning with a revisiting and introspective behavior. Our experiments demonstrate that, under the same annotation budget, our method is more data-efficient and yields a relatively noise-free annotation dataset in the end.", "date": "2026-06-07", "categories": []}
{"id": "2603.21180", "title": "ALMAB-DC: Active Learning, Multi-Armed Bandits, and Distributed Computing for Sequential Experimental Design and Black-Box Optimization", "abstract": "Sequential experimental design under expensive, gradient-free objectives is a central challenge in computational statistics: evaluation budgets are tightly constrained and information must be extracted efficiently from each observation. We propose \\textbf{ALMAB-DC}, a GP-based sequential design framework combining active learning, multi-armed bandits (MAB), and distributed asynchronous computing for expensive black-box experimentation. A Gaussian process surrogate with uncertainty-aware acquisition identifies informative query points; a UCB or Thompson-sampling bandit controller allocates evaluations across parallel workers; and an asynchronous scheduler handles heterogeneous runtimes. We present cumulative regret bounds for the bandit components and characterize parallel scalability via Amdahl's Law.\n  We validate ALMAB-DC on five benchmarks. On the two statistical experimental-design tasks, ALMAB-DC achieves lower simple regret than Equal Spacing, Random, and D-optimal designs in dose--response optimization, and in adaptive spatial field estimation matches the Greedy Max-Variance benchmark while outperforming Latin Hypercube Sampling; at $K=4$ the distributed setting reaches target performance in one-quarter of sequential wall-clock rounds. On three ML/engineering tasks (CIFAR-10 HPO, CFD drag minimization, MuJoCo RL), ALMAB-DC achieves 93.4\\% CIFAR-10 accuracy (outperforming BOHB by 1.7\\,pp and Optuna by 1.1\\,pp), reduces airfoil drag to $C_D = 0.059$ (36.9\\% below Grid Search), and improves RL return by 50\\% over Grid Search. All advantages over non-ALMAB baselines are statistically significant under Bonferroni-corrected Mann--Whitney $U$ tests. Distributed execution achieves $7.5\\times$ speedup at $K = 16$ agents, consistent with Amdahl's Law.", "date": "2026-03-22", "categories": []}
{"id": "2606.04647", "title": "ALINC: Active Learning for Inductive Node Classification via Graph Sampling", "abstract": "Active learning (AL) for node classification typically focuses on selecting the most informative nodes for annotation within one or a few large graphs (e.g., in social network analysis). However, in other domains, such as molecular chemistry or electronic design automation, datasets consist of thousands of independent graphs. In many of these inductive settings, annotating an individual node requires a full-graph analysis, which effectively yields the remaining node labels on-the-fly. Therefore, these scenarios require AL strategies that select entire graphs instead of single nodes, a problem which has not been tackled in the literature so far. Thus, we introduce ALINC, an AL framework for inductive node classification via graph sampling. It bridges the existing methodological gap by elevating node-level utility measures to graph-level selection criteria through various aggregation mechanisms. In an extensive benchmark including ten strategies, three aggregation methods, and four datasets, we identify CoreSet, TypiClust, and BADGE as the top-performing graph sampling strategies. Our detailed analysis further reveals that the choice of the aggregation method is pivotal, as it substantially affects model performance and annotation costs. Finally, we demonstrate the effectiveness of ALINC in two use case studies: site-of-metabolism prediction in molecules and design automation of printed circuit board schematics.", "date": "2026-06-03", "categories": []}
{"id": "2606.05134", "title": "Activation-Based Active Learning for In-Context Learning: Challenges and Insights", "abstract": "Deep active learning has previously been explored for LLM in-context sample selection, but not with methods that utilise recent advances in understanding of transformer activations. In this paper, we test the hypothesis that model activations could provide a fine-grained signal to optimise the selection of in-context examples. We present the most comprehensive analysis to date of MLP activation-based deep active learning methods applied to in-context learning, including how different attention masking strategies impact active learning across diverse classification and generative datasets, using both Llama-3.2-3B and Qwen2.5-3B base models. However, we find a negative result: MLP outputs, viewed through the lenses of massive activations or the first four moments, do not correlate with example quality or task performance. Specifically, the absolute Spearman correlation coefficient is at most 0.33 for all tasks and models we tested, showing that such activation-based sampling should not be used for in-context learning. We hypothesise that this may be due to superposition, whereby models represent more features than they have dimensionality, suggesting that methods like Sparse Autoencoders (SAEs) may be a promising future direction.", "date": "2026-06-03", "categories": []}
{"id": "2606.03821", "title": "Finding Needles in the Haystack: Transductive Active Labeling in Ecology", "abstract": "Active learning is now standard practice in labeling ecological data, enabling ecologists to quickly process large volumes of field data to understand and monitor natural environments. Current practices evaluate active learning inductively, estimating predictive performance on a held-out test set. We argue that this evaluation is misaligned with most ecological tasks, where the goal is to transductively label an entire pool of data as efficiently as possible. We demonstrate that ignoring the human-in-the-loop underestimates the importance of continuing to label, particularly for classes in the long tail which may be of disproportionate ecological importance (rare species, uncommon behaviors, etc.). Our analysis shows that, for this long tail, the transductive objective shifts importance from prediction to discovery: the true challenge becomes finding \"needles in the haystack,\" examples of rare classes that are embedded within dense regions of abundant classes in the latent geometry, which we quantify with a novel metric of sampling difficulty. Finally, to translate these insights to practical ecological workflows, we propose a conservative hybrid stopping criterion inspired by ecological rarefaction curves, and show that combining predictive performance with discovery criteria reduces premature stopping on long-tailed pools, improving rare-class recovery when discovery, not classification, is the limiting factor.", "date": "2026-06-02", "categories": []}
{"id": "2311.00260", "title": "Incentivized Collaboration in Active Learning", "abstract": "In collaborative active learning, where multiple agents try to learn labels from a common hypothesis, we introduce an innovative framework for incentivized collaboration. Here, rational agents aim to obtain labels for their data sets while keeping label complexity at a minimum. We focus on designing (strict) individually rational (IR) collaboration protocols, ensuring that agents cannot reduce their expected label complexity by acting individually. We first show that given any optimal active learning algorithm, the collaboration protocol that runs the algorithm as is over the entire data is already IR. However, computing the optimal algorithm is NP-hard. We therefore provide collaboration protocols that achieve (strict) IR and are comparable with the best known tractable approximation algorithm in terms of label complexity.", "date": "2023-11-01", "categories": []}
{"id": "2510.12624", "title": "Learning-To-Measure: In-Context Active Feature Acquisition", "abstract": "Active feature acquisition (AFA) is a sequential decision-making problem where the goal is to improve model performance for test instances by adaptively selecting which features to acquire. In practice, AFA methods often learn from retrospective data with systematic missingness in the features and limited task-specific labels. Most prior work addresses acquisition for a single predetermined task, limiting scalability. To address this limitation, we formalize the meta-AFA problem, where the goal is to learn acquisition policies across various tasks. We introduce Learning-to-Measure (L2M), which consists of i) reliable uncertainty quantification over unseen tasks, and ii) an uncertainty-guided greedy feature acquisition agent that maximizes conditional mutual information. We demonstrate a sequence-modeling or autoregressive pre-training approach that underpins reliable uncertainty quantification for tasks with arbitrary missingness. L2M operates directly on datasets with retrospective missingness and performs the meta-AFA task in-context, eliminating per-task retraining. Across synthetic and real-world tabular benchmarks, L2M matches or surpasses task-specific baselines, particularly under scarce labels and high missingness.", "date": "2025-10-14", "categories": []}
{"id": "2606.02079", "title": "FACT: A Simple and Efficient Framework for Active Finetuning", "abstract": "The main goal of active finetuning is to improve a pretrained model's performance on a specific task or domain by finetuning it with carefully selected informative or challenging data. Previous research has predominantly focused on the active aspect (i.e., data selection) while uniformly employing full finetuning for model adaptation, which inevitably distorts pretrained features due to distribution shift. This issue becomes particularly pronounced when the model size is large relative to the finetuning data quantity, leading to heightened overfitting risks. To address this critical gap, we formally outline the FiAF task that emphasizes systematic exploration of finetuning methodologies in active learning. We propose FACT, a three-phase hierarchical finetuning framework featuring both efficiency and simplicity, specifically designed for active finetuning scenarios. Our comprehensive experiments span: (1) Three major dataset categories encompassing classic (CIFAR10, CIFAR100, ImageNet-1k), imbalanced (CIFAR10-LT, CIFAR100-LT), and fine-grained (StanfordCars, FGVCAircraft) image classification datasets, each evaluated under 3-5 distinct sampling ratios; (2) Diverse pretrained architectures including Convolutional Neural Network (ConvNeXt), Vision Transformer (ViT), and Vision LSTM (ViL) networks; (3) A systematic investigation of frozen feature augmentation (FroFA) strategies. (4) A comprehensive and rigorous analysis of efficiency and generalizability. The results demonstrate significant improvements with strong generalization and robustness. Notably, under low sampling ratios, our framework achieves remarkable performance gains of over 20% on the ViT model for CIFAR10, CIFAR100, and ImageNet-1k benchmarks. This systematic approach establishes new state-of-the-art performance while maintaining parameter efficiency, proving particularly effective when labeled data is scarce.", "date": "2026-06-01", "categories": []}
{"id": "2605.30625", "title": "Active Timepoint Selection for Learning Measure-Valued Trajectories", "abstract": "Inferring continuous probability paths from sparse snapshots is a fundamental challenge in domains like single-cell biology, where high-fidelity data acquisition is often destructive and constrained by prohibitive sequencing costs. This motivates the need for active learning strategies to strategically select optimal measurement times. However, designing active learning policies for this setting remains an open problem: the target objects reside on the infinite dimensional Wasserstein space where standard Euclidean metrics are ill-defined, and current interpolation methods lack epistemic uncertainty quantification. We introduce a framework which extends active experimentation to the space of measures. By leveraging Linearized Optimal Transport (LOT), we map distributional snapshots into a tangent space amenable to Gaussian Process modeling, allowing us to construct a tractable probabilistic surrogate for the underlying probability path. This yields an acquisition policy that iteratively selects measurement times to minimize uncertainty. Empirical results demonstrate that our strategy outperforms uncertainty-agnostic baselines on both synthetic and real-world datasets.", "date": "2026-05-28", "categories": []}
{"id": "2602.17605", "title": "Adapting Actively on the Fly: Relevance-Guided Online Meta-Learning with Latent Concepts for Geospatial Discovery", "abstract": "In environmental monitoring, data collection is often costly, sparse, and shaped by urgent public-health needs. This is particularly true for cancer-causing PFAS (Per- and polyfluoroalkyl substances) contamination, where discussions with domain experts and environmental organizations highlight the need to strategically identify high-risk, under-observed regions under tight sampling budgets. More broadly, similar challenges arise in disaster response and public health settings, where dynamic environments make it essential to efficiently uncover hidden targets from limited ground truth. Yet sparse and biased geospatial labels limit the applicability of existing learning-based methods, such as reinforcement learning. To address this, we propose a unified geospatial discovery framework that integrates active learning, online meta-learning, and concept-guided reasoning. Our approach introduces two key innovations built on a shared notion of *concept relevance*, capturing how domain-specific factors influence target presence: a *concept-weighted uncertainty sampling strategy*, where uncertainty is modulated by learned relevance from readily available concepts such as land cover and source proximity; and a *relevance-aware meta-batch formation strategy* that promotes semantic diversity during online-meta updates, improving generalization in dynamic environments. We evaluate our framework on PFAS contamination discovery as a real-world inspired environmental monitoring task, demonstrating robust target discovery under limited data and changing conditions.", "date": "2026-02-19", "categories": []}
{"id": "2605.26072", "title": "Active Query Synthesis for Preference Learning", "abstract": "Efficient learning of user preferences is crucial for many modern decision making systems but typically requires costly labeled data. Active learning reduces this cost, yet standard methods are computationally expensive due to pool-based evaluation. Further, most methods assume all query feedback is equally reliable, ignoring that pairwise queries between nearly identical or entirely dissimilar items yield ambiguous, low-confidence responses. To address the issue of feedback reliability, we introduce a novel confidence aware response model that explicitly accounts for these ambiguous comparisons. To overcome the computational bottleneck of pool-based evaluation, we propose an active query synthesis framework, Info-Synth that generates optimal queries by maximizing a mutual information-based objective within a continuous space. Moreover, we propose two strategies, Pair M-dist and Pair Opt-dist, that extend Info-Synth to select effective queries even when restricted to finite query pools. We demonstrate our framework's versatility and performance across synthetic preference learning, constrained text summary datasets, and subjective, continuous-space controller gain tuning for a simulated mobile robot.", "date": "2026-05-25", "categories": []}
{"id": "2509.26005", "title": "BALLAST: Bayesian Active Learning with Look-ahead Amendment for Sea-drifter Trajectories under Spatio-Temporal Vector Fields", "abstract": "We introduce a formal active learning methodology for guiding the placement of Lagrangian observers to infer time-dependent vector fields -- a key task in oceanography, marine science, and ocean engineering -- using a physics-informed spatio-temporal Gaussian process surrogate model. The majority of existing placement campaigns either follow standard `space-filling' designs or relatively ad-hoc expert opinions. A key challenge to applying principled active learning in this setting is that Lagrangian observers are continuously advected through the vector field, so they make measurements at different locations and times. It is, therefore, important to consider the likely future trajectories of placed observers to account for the utility of candidate placement locations. To this end, we present BALLAST: Bayesian Active Learning with Look-ahead Amendment for Sea-drifter Trajectories. We observe noticeable benefits of BALLAST-aided sequential observer placement strategies on both synthetic and high-fidelity ocean current models. In addition, we developed a novel GP inference method -- the Vanilla SPDE Exchange (VaSE) -- to boost the GP posterior sampling efficiency, which is also of independent interest.", "date": "2025-09-30", "categories": []}
{"id": "2605.22169", "title": "Balancing Uncertainty and Diversity of Samples: Leveraging Diversity of Least, High Confidence Samples for Effective Active Learning", "abstract": "Deep learning models, including Convolutional Neural Networks (CNNs) and Vision Transformers (ViTs), have achieved state-of-the-art performance on various computer vision tasks such as object classification, detection, segmentation, generation, and many more. However, these models are data-hungry as they require more training data to learn millions or billions of parameters. Especially for supervised learning tasks, curating a large number of labeled samples for model training is an expensive and time-consuming task. Active Learning (AL) has been used to address this problem for many years. Existing active learning methods aim at choosing the samples for annotation from a pool of unlabeled samples that are either diverse or uncertain. Choosing such samples may hinder the model's performance as we pool based on one dimension, i.e., either diverse or uncertain. In this paper, we propose four novel hybrid sampling methods for pooling both easy and hard samples, which are also diverse. To verify the efficacy of the proposed methods, extensive experiments are conducted using high and low-confidence samples separately. We observe from our experiments that the proposed hybrid sampling method, Least Confident and Diverse (LCD), consistently performs better compared to state-of-the-art methods. It is observed that selecting uncertain and diverse instances helps the model learn more distinct features. The codes related to this study will be available at https://github.com/XXX/LCD.", "date": "2026-05-21", "categories": []}
{"id": "2605.20495", "title": "A Human-in-the-Loop Framework for Efficient Prompt Selection in Microscopy Vision-Language Models", "abstract": "Deep-learning pipelines for microscopy image classification often require expensive, labor- and time-intensive expert annotation to produce high-quality ground truth for training. Recent work has shown that prompt tuning of vision-language models (VLMs) can reduce manual annotation by constructing a small prompt set of expert-verified image-caption exemplars that is reused as few-shot context to classify all remaining images at inference time. To further reduce effort, the VLM can draft captions for candidate exemplars, which experts then verify and lightly edit instead of writing text de novo. However, two practical questions remain unaddressed: (1) which unlabeled images should be prioritized for verification, and (2) how many verified exemplars are needed to reach a performance target. In this work, we address these questions by formulating prompt-set construction as a target-driven active learning problem that prioritizes which images to annotate. We study three complementary selection criteria under strict low-resource constraints with small unlabeled pools. Experiments show that our methods reach the target performance with substantially fewer expert-verified images than random selection, achieving 100% test accuracy with as few as 20 annotated images on average. More broadly, our human-in-the-loop framework demonstrates a human-centered use of generative AI in biomedical image analysis, where experts remain actively involved in verifying and refining model output while significantly reducing annotation cost. Code and data will be publicly available.", "date": "2026-05-19", "categories": []}
{"id": "2605.20771", "title": "Cumulative Meta-Learning from Active Learning Queries for Robustness to Spurious Correlations", "abstract": "Spurious correlations in real-world datasets cause machine learning models to rely on irrelevant patterns, undermining reliability, generalization, and fairness. Active learning offers a promising way to address this failure mode by querying informative samples that distinguish core features from spurious ones. However, standard active-learning methods simply append queried examples to the labeled set, effectively updating only the likelihood term. In deep learning regimes, the influence of these informative samples can be diluted by the larger labeled set and memorized by overparameterized models. We propose Cumulative Active Meta-Learning (CAML), an active-learning framework that uses queried examples to meta-learn the prior, or inductive bias, governing how the model adapts. CAML casts each active-learning round as a meta-learning task: the current labeled set serves as meta-train data for adaptation, while the newly queried batch serves as meta-test data for evaluating generalization. Unlike conventional meta-learning, which treats tasks as independent and identically distributed, CAML exploits the sequential dependence between active-learning rounds by maintaining a cumulative inductive bias that is progressively refined. Theoretically, we show that this cumulative formulation introduces interaction terms that couple earlier meta-learned inductive biases with later query-induced objectives, capturing dependencies absent from standard meta-learning. Empirically, CAML improves minority-group accuracy across spurious-correlation benchmarks and acquisition strategies, with gains of up to 27.8% on Dominoes, 29.9% on Waterbirds, 14.3% on SpuCo, and 24.0% on CivilComments.", "date": "2026-05-20", "categories": []}
{"id": "2605.02609", "title": "Gradient-Discrepancy Acquisition for Pool-Based Active Learning", "abstract": "The effectiveness of active learning hinges on the choice of the acquisition criterion by which a learning algorithm selects potentially informative data points whose label is subsequently queried. This paper proposes a novel gradient-based acquisition criterion, derived from a generalization bound introduced by Luo et al. (2022). This criterion can be applied in lieu of uncertainty measures in uncertainty sampling, or incorporated into diversity-based methods that consider the spread of sampled points in addition to the uncertainty of their labels. We provide a theoretical justification of the proposed acquisition criterion, and demonstrate its effectiveness in an empirical evaluation.", "date": "2026-05-04", "categories": []}
{"id": "2605.03964", "title": "Pretrained Model Representations as Acquisition Signals for Active Learning of MLIPs", "abstract": "Training machine learning interatomic potentials (MLIPs) for reactive chemistry is often bottlenecked by the high cost of quantum chemical labels and the scarcity of transition state configurations in candidate pools. Active learning (AL) can mitigate these costs, but its effectiveness hinges on the acquisition rule. We investigate whether the latent space of a pretrained MLIP already contains the information necessary for effective acquisition, eliminating the need for auxiliary uncertainty heads, Bayesian training and fine-tuning, or committee ensembles. We introduce two acquisition signals derived directly from a pretrained MACE potential: a finite-width neural tangent kernel (NTK) and an activation kernel built from hidden latent space features. On reactive-chemistry benchmarks, both kernels consistently outperform fixed-descriptor baselines, committee disagreement, and random acquisition, reducing the data required to reach performance targets by an average of 38% for energy error and 28% for force error. We further show that the pretrained model induces similarity spaces that preserve chemically meaningful structure and provide more reliable residual uncertainty estimates than randomly initialised or fixed-descriptor-based kernels. Our results suggest that pretraining aligns latent-space geometry with model error, yielding a practical and sufficient acquisition signal for reactive MLIP fine-tuning.", "date": "2026-05-05", "categories": []}
{"id": "2605.13788", "title": "Force-Aware Neural Tangent Kernels for Scalable and Robust Active Learning of MLIPs", "abstract": "Active learning for machine-learning interatomic potentials (MLIPs) must address several challenges to be practical: scaling to large candidate pools, leveraging energy-force supervision, and maintaining robustness when candidate pools are biased relative to the target distribution. In this work, we jointly address these challenges. We first introduce a linearly scaling acquisition framework based on chunked feature-space posterior-variance shortlisting. By avoiding materialisation of the candidate and train set kernels, this approach enables screening of ~200k structures within hours and applies broadly to acquisition strategies that score candidates based on molecular similarity metrics. We then extend the Neural Tangent Kernel (NTK) to a force-aware setting via mixed parameter-coordinate derivatives, yielding a force NTK and a joint energy-force NTK that provide natural similarity metrics for vector-field prediction. We demonstrate the effectiveness of the joint energy-force NTK on the OC20 dataset, where force-aware acquisition is crucial: it achieves the lowest energy and force MAE and RMSE across all metrics and distribution splits. Across T1x, PMechDB, and RGD benchmarks, our force NTK methods remain competitive with established baselines while being significantly more efficient than committee-based approaches. Under a controlled candidate-pool shift case study on T1x, acquisition based on pretrained MLIP embeddings and NTKs remains robust, whereas committee-based methods exhibit higher variance. Overall, these results show that a single pretrained MLIP can enable scalable, force-aware, and distribution-robust active learning for foundation-model fine-tuning.", "date": "2026-05-13", "categories": []}
{"id": "2605.14689", "title": "Are Candidate Models Really Needed for Active Learning?", "abstract": "Deep learning has profoundly impacted domains such as computer vision and natural language processing by uncovering complex patterns in vast datasets. However, the reliance on extensive labeled data poses significant challenges, including resource constraints and annotation errors, particularly in training Convolutional Neural Networks (CNNs) and transformers due to a larger number of parameters. Active learning offers a promising solution to reduce labeling burdens by strategically selecting the most informative samples for annotation. However, the current active learning frameworks are time-intensive which select the samples iteratively with the help of initial candidate models. This study investigates the feasibility of using CNNs and transformers with randomly initialized weights, eliminating the need for initial candidate models while achieving results comparable to active learning frameworks that depend on such candidate models. We evaluate three confidence-based sampling strategies: high confidence (HC), low confidence (LC), and a combination of high confidence in the early stages of training and low confidence at later stages of training (HCLC). Among these, mostly LC demonstrated the best performance in our experiments, showcasing its effectiveness as an active learning strategy without the need for candidate models. Further, extensive experiments verify the robustness of the proposed active learning methods. By challenging traditional frameworks, the proposed work introduces a streamlined approach to active learning, advancing efficiency and flexibility across diverse datasets and domains.", "date": "2026-05-14", "categories": []}
{"id": "2605.14917", "title": "A Mutual Information Lower Bound for Multimodal Regression Active Learning", "abstract": "Active learning for continuous regression has lacked an acquisition function that targets epistemic uncertainty when the predictive distribution is multimodal: variance misses modal disagreement, and information-theoretic targets like BALD are designed for discrete outputs. We introduce a Two-Index framework that makes this separation explicit: one stochastic index selects among competing model hypotheses (epistemic source), while a second governs within-hypothesis randomness (aleatoric source). An entropy decomposition within the framework identifies the mutual information between the output and the epistemic index as a principled acquisition objective, and we prove this quantity vanishes as the model is trained on growing datasets, confirming that it captures exactly the uncertainty data can resolve. Because this mutual information is intractable for continuous outputs, we derive the Mutual Information Lower Bound (MI-LB) acquisition function, a closed-form approximation for Mixture Density Network ensembles. On benchmarks featuring multimodal systems, MI-LB matches or beats every baseline evaluated and is the only method to do so consistently -- geometric and Fisher-based baselines compete only when the input space already encodes the multimodality, and collapse otherwise.", "date": "2026-05-14", "categories": []}
{"id": "2605.11638", "title": "Learning U-Statistics with Active Inference", "abstract": "$U$-statistics play a central role in statistical inference. In many modern applications, however, acquiring the labels required for $U$-statistics is costly. Motivated by recent advances in active inference, we develop an active inference framework for $U$-statistics that selectively queries informative labels to improve estimation efficiency under a fixed labeling budget, while preserving valid statistical inference. Our approach is built on the augmented inverse probability weighting $U$-statistic, which is designed to incorporate the sampling rule and machine learning predictions. We characterize the optimal sampling rule that minimizes its variance and design practical sampling strategies. We further extend the framework to $U$-statistic-based empirical risk minimization. Experiments on real datasets demonstrate substantial gains in estimation efficiency over baseline methods, while maintaining target coverage.", "date": "2026-05-12", "categories": []}
{"id": "2506.20928", "title": "Active Learning for Manifold Gaussian Process Regression", "abstract": "This paper introduces an active learning framework for manifold Gaussian Process (GP) regression, combining manifold learning with strategic data selection to improve accuracy in high-dimensional spaces. Our method jointly optimizes a neural network for dimensionality reduction and a Gaussian process regressor in the latent space, supervised by an active learning criterion that minimizes global prediction error. Experiments on synthetic data demonstrate superior performance over randomly sequential learning. The framework efficiently handles complex, discontinuous functions while preserving computational tractability, offering practical value for scientific and engineering applications. Future work will focus on scalability and uncertainty-aware manifold learning.", "date": "2025-06-26", "categories": []}
{"id": "2510.09877", "title": "Batch Bayesian Active Learning with Partial Batch Label Sampling", "abstract": "Over the past couple of decades, many active learning acquisition functions have been proposed, leaving practitioners with an unclear choice of which to use. Bayesian-based active learning offers principled objectives with explainable intuition, including Expected Error Reduction (EER), Expected Predictive Information Gain (EPIG), and Bayesian Active Learning by Disagreements (BALD). A key challenge of such methods is the difficult scaling to large batch sizes, leading to either computational challenges (BatchBALD) or dramatic performance drops (top-$B$ selection). Here, using a particular formulation of Bayesian Decision Theory, we derive Partial Batch Label Sampling (ParBaLS) for the EPIG algorithm. We show experimentally for several datasets that ParBaLS EPIG gives superior performance for a fixed budget and Bayesian Logistic Regression on embeddings from large pre-trained models. Our code is available at https://github.com/ADDAPT-ML/ParBaLS.", "date": "2025-10-10", "categories": []}
{"id": "2603.13536", "title": "Active Sampling Sample-based Quantum Diagonalization from Finite-Shot Measurements", "abstract": "Near-term quantum devices provide only finite-shot measurements and prepare imperfect, contaminated states. This motivates algorithms that convert samples into reliable low-energy estimates without full tomography or exhaustive measurements. We propose Active Sampling Sample-based Quantum Diagonalization (AS-SQD), framing SQD as an active learning problem: given measured bitstrings, which additional basis states should be included to efficiently recover the ground-state energy? SQD restricts the Hamiltonian to a selected set of basis states and classically diagonalizes the restricted matrix. However, naive SQD using only sampled states suffers from bias under finite-shot sampling and excited-state contamination, while blind random expansion is inefficient as system size grows. We introduce a perturbation-theoretic acquisition function based on Epstein--Nesbet second-order energy corrections to rank candidate basis states connected to the current subspace. At each iteration, AS-SQD diagonalizes the restricted Hamiltonian, generates connected candidates, and adds the most valuable ones according to this score. We evaluate AS-SQD on disordered Heisenberg and Transverse-Field Ising (TFIM) spin chains up to 16 qubits under a preparation model mixing 80\\% ground state and 20\\% first excited state. Furthermore, we validate its robustness against real-world state preparation and measurement (SPAM) errors using physical samples from an IBM Quantum processor. Across simulated and hardware evaluations, AS-SQD consistently achieves substantially lower absolute energy errors than standard SQD and random expansion. Detailed ablation studies demonstrate that physics-guided basis acquisition effectively concentrates computation on energetically relevant directions, bypassing exponential combinatorial bottlenecks.", "date": "2026-03-13", "categories": []}
{"id": "2605.08429", "title": "Active Multiple-Prediction-Powered Inference", "abstract": "Post-deployment monitoring of healthcare AI requires statistically valid, label-efficient methods, but gold-standard labels from clinician chart review are expensive. Prediction-powered inference (PPI) and active statistical inference (ASI) reduce label cost by combining a small labeled sample with abundant model predictions, but both are restricted to a single predictor, a poor fit for modern clinical pipelines that have multiple predictors of differing cost and accuracy available at inference time. We propose Active Multiple-Prediction-Powered Inference (AM-PPI), which routes each instance to a cost-appropriate predictor subset, samples gold-standard labels in proportion to the chosen subset's residual uncertainty, and reweights predictions to minimize estimator variance, all under a single deployment-time budget. AM-PPI generalizes ASI to leverage multiple predictors and extends Multiple-PPI from global per-predictor allocation to per-instance adaptive routing. We derive closed-form Karush-Kuhn-Tucker (KKT) conditions for all three decisions and prove, via biconvexity and strong duality, that the resulting fixed point is a global optimum despite the joint problem being non-jointly-convex. We establish asymptotic normality with valid coverage, minimum-variance unbiasedness within the linear-prediction augmented inverse propensity weighted (AIPW) class, and a closed-form criterion identifying when multiple predictors help. On synthetic data and three healthcare monitoring tasks, AM-PPI produces 10 to 40 percent narrower confidence intervals (CIs) than single-predictor ASI in the budget regime where routing matters, and matches the better baseline elsewhere.", "date": "2026-05-08", "categories": []}
{"id": "2605.09858", "title": "Clip-level Uncertainty and Temporal-aware Active Learning for End-to-End Multi-Object Tracking", "abstract": "Multi-Object Tracking (MOT) in dynamic environments relies on robust temporal reasoning to maintain consistent object identities over time. Transformer-based end-to-end MOT models achieve strong performance by explicitly modeling temporal dependencies, yet training them requires extensive bounding-box and identity annotations. Given the high labeling cost and strong redundancy in videos, Active Learning (AL) is an effective approach to improve annotation efficiency. However, existing AL methods for MOT primarily operate at the frame level, which is structurally misaligned with modern end-to-end trackers whose inference and training rely on multi-frame clips. To bridge this gap, we formulate clip-level active learning and propose Clip-level Uncertainty and Temporal-aware Active Learning (CUTAL). In contrast to frame-based approaches, CUTAL scores each clip using uncertainty metrics derived from multi-frame predictions to capture inter-frame correspondence ambiguities, while enforcing temporal diversity to select an informative and non-redundant subset. Experiments show that CUTAL achieves stronger overall performance than baselines at the same label budgets across MeMOTR and SambaMOTR. Notably, CUTAL achieves performance comparable to full supervision for MeMOTR on both datasets using only 50% of the labeled training data.", "date": "2026-05-11", "categories": []}
{"id": "2605.10349", "title": "Portable Active Learning for Object Detection", "abstract": "Annotating bounding boxes is costly and limits the scalability of object detection. This challenge is compounded by the need to preserve high accuracy while minimizing manual effort in real-world applications. Prior active learning methods often depend on model features or modify detector internals and training schedules, increasing integration overhead. Moreover, they rarely jointly exploit the benefits of image-level signals, class-imbalance cues, and instance-level uncertainty for comprehensive selection. We present Portable Active Learning (PAL), a detector-agnostic, easily portable framework that operates solely on inference outputs. PAL combines class-wise instance uncertainty with image-level diversity to guide data selection. At each round, PAL trains lightweight class-specific logistic classifiers to distinguish true from false positives, producing entropy-based uncertainty scores for proposals. Candidate images are then refined using global image entropy, class diversity, and image similarity, yielding batches that are both informative and diverse. PAL requires no changes to model internals or training pipelines, ensuring broad compatibility across detectors. Extensive experiments on COCO, PASCAL VOC, and BDD100K demonstrate that PAL consistently improves label efficiency and detection accuracy compared to existing active learning baselines, making it a practical solution for scalable and cost-effective deployment of object detection in real-world settings.", "date": "2026-05-11", "categories": []}
{"id": "2605.10654", "title": "Active Learning for Gaussian Process Regression Under Self-Induced Boltzmann Weights", "abstract": "We consider the active learning problem where the goal is to learn an unknown function with low prediction error under an unknown Boltzmann distribution induced by the function itself. This self-induced weighting arises naturally in problems such as potential energy surface (PES) modeling in computational chemistry,\n  yet poses unique challenges as the target distribution is unknown and its partition function is intractable. We propose \\texttt{AB-SID-iVAR}, a Gaussian Process-based acquisition function that approximates the intractable Bayesian target distribution in closed form while avoiding partition function estimation, and is applicable to both discrete and continuous input domains. We also analyze a Thompson sampling alternative (\\texttt{TS-SID-iVAR}) as a higher variance Monte Carlo variant. Despite the unknown target, under mild conditions, we establish that the terminal prediction error vanishes with high probability, and provide a tighter average-case guarantee. We demonstrate consistent improvements over existing approaches in this setting on synthetic benchmarks and real-world PES modeling and drug discovery tasks.", "date": "2026-05-11", "categories": []}
{"id": "2604.11995", "title": "Loss-Driven Bayesian Active Learning", "abstract": "The central goal of active learning is to gather data that maximises downstream predictive performance, but popular approaches have limited flexibility in customising this data acquisition to different downstream problems and losses. We propose a rigorous loss-driven approach to Bayesian active learning that allows data acquisition to directly target the loss associated with a given decision problem. In particular, we show how any loss can be used to derive a unique objective for optimal data acquisition. Critically, we then show that any loss taking the form of a weighted Bregman divergence permits analytic computation of a central component of its corresponding objective, making the approach applicable in practice. In regression and classification experiments with a range of different losses, we find our approach reduces test losses relative to existing techniques.", "date": "2026-04-13", "categories": []}
{"id": "2605.06756", "title": "Physics-based Digital Twins for Integrated Thermal Energy Systems Using Active Learning", "abstract": "Real-time supervisory control of thermal energy distribution systems requires digital twins that are accurate, interpretable, and uncertainty-aware, yet remain data and computationally efficient. High-fidelity simulations alone are costly, while purely data-driven surrogates often lack robustness. To address these challenges, this work proposes an active learning (AL) framework that couples system-level Modelica simulations with four simpler physics-informed and data-driven surrogate modeling approaches: deterministic Sparse Identification of Nonlinear Dynamics with Control (SINDyC), its probabilistic multivariate-Gaussian extension (MvG-SINDyC), feedforward neural network (FNN), and gated recurrent unit (GRU) network. Tailored to each surrogate, model-specific AL query strategies are employed, including Mahalanobis-distance sampling in coefficient space for MvG-SINDyC and error-based sampling in prediction space for SINDyC, FNN, and GRU, allowing the learning process to prioritize dynamically informative trajectories.\n  The proposed approach is demonstrated on the glycol heat exchanger (GHX) subsystem of the Thermal Energy Distribution System (TEDS) at Idaho National Laboratory. Across key GHX outputs--the bypass mass flow rate $\\dot{m}_{\\mathrm{GHX}}$ and heat transfer rate $Q_{\\mathrm{GHX}}$-the AL framework achieves comparable predictive accuracy using as few as one-fifth of the simulation trajectories required by random sampling. Among the evaluated surrogates, the GRU achieves the highest predictive fidelity, while SINDyC remains the most computationally efficient and interpretable. The probabilistic MvG-SINDyC surrogate further enables uncertainty quantification and exhibits the largest computational gains under AL.", "date": "2026-05-07", "categories": []}
{"id": "2605.07758", "title": "SMT-Based Active Learning of Weighted Automata", "abstract": "We present an SMT-based active learning algorithm for nondeterministic weighted automata (WFAs) as a practical and robust alternative to Hankel/L*-style methods. Our algorithm is parametric in a given semiring and, if it terminates, guaranteed to produce minimal WFAs. We prove partial correctness and provide a sufficient termination condition, which in particular implies termination for all finite semirings. Our extensive experimental evaluation shows that our algorithm is capable of learning numerous minimal WFAs over both finite and infinite semirings, vastly outperforms a naive baseline, and is competitive with a state-of-the-art algorithm while producing significantly smaller automata and requiring less interaction with the teacher.", "date": "2026-05-08", "categories": []}
{"id": "2605.05511", "title": "Non-Myopic Active Feature Acquisition via Pathwise Policy Gradients", "abstract": "Active feature acquisition (AFA) considers prediction problems in which features are costly to obtain and the learner adaptively decides which feature values to acquire for each instance and when to stop and predict. AFA can be formulated as a partially observable Markov decision process (POMDP), which naturally admits a sequential decision-making perspective. In this paper, we present non-myopic pathwise policy gradients (NM-PPG), a new AFA method built around this formulation. We introduce a continuous relaxation of the acquisition process that enables pathwise gradients through the full acquisition trajectory, avoiding the high variance of standard score-function policy gradients while allowing end-to-end optimization of a non-myopic acquisition policy. To better align training with deployment, we further develop a straight-through rollout scheme that follows hard feature acquisitions in the forward pass while backpropagating through the corresponding soft relaxation in the backward pass. We stabilize optimization with entropy regularization and staged temperature sharpening. Experiments on both synthetic and real-world datasets demonstrate that NM-PPG yields superior performance relative to state-of-the-art AFA baselines.", "date": "2026-05-06", "categories": []}
{"id": "2605.00868", "title": "Autonomous Reliability Qualification of Ga$_2$O$_3$-based Hydrogen and Temperature Sensors via Safe Active Learning", "abstract": "We present a Safe Active Learning (SAL) framework for autonomous reliability characterization of rectifying Ga$_2$O$_3$-based devices under coupled thermal and hydrogen stress. SAL treats rectification as a device-physics-motivated safety observable and models its evolution over elapsed time, temperature, and H$_2$ concentration using a Gaussian-process surrogate. To handle condition-dependent and uncertain experiment durations, the method combines an adaptive completion-time window, time-window lower-confidence-bound safety checks, a trust region anchored to previously verified safe conditions, and a two-phase strategy that transitions from conservative safe exploration to progressively relaxed rectification targets as the device degrades. We first evaluate SAL in simulation, where it safely expands the explored region while learning the evolving rectification surface. We then demonstrate SAL experimentally on an automated high-temperature probe-station platform using a Pt/Cr$_2$O$_3$:Mg/$\\beta$-Ga$_2$O$_3$ device. In the reported campaign, phase 1 incurred only one unsafe measurement associated with spurious current-voltage sweeps, while phase 2 intentionally probed lower-rectification regimes. Finally, we use the curated SAL dataset for offline long-horizon forecasting of device response at a target voltage using a structured Gaussian-process model with a condition-dependent Kohlrausch--Williams--Watts mean and a residual covariance kernel. The model captures long-time, saturating degradation trends in an auxiliary validation dataset, illustrating how safety-aware autonomous experimentation enables both conservative characterization and subsequent degradation modeling. Although demonstrated here for a rectifying Ga$_2$O$_3$ device, SAL is applicable to other systems where a measurable in situ safety observable can be defined.", "date": "2026-04-22", "categories": []}
{"id": "2604.02019", "title": "Feature Weighting Improves Pool-Based Sequential Active Learning for Regression", "abstract": "Pool-based sequential active learning for regression (ALR) optimally selects a small number of samples sequentially from a large pool of unlabeled samples to label, so that a more accurate regression model can be constructed under a given labeling budget. Representativeness and diversity, which involve computing the distances among different samples, are important considerations in ALR. However, previous ALR approaches do not incorporate the importance of different features in inter-sample distance computation, resulting in inaccurate distances and hence sub-optimal sample selection. This paper proposes four feature weighted single-task ALR approaches and three feature weighted multi-task ALR approaches, where the ridge regression coefficients trained from a small amount of previously labeled samples are used to weight the corresponding features in inter-sample distance computation. Extensive experiments showed that this intuitive and easy-to-implement enhancement almost always improves the performance of five existing ALR approaches, in both single-task and multi-task regression problems. The feature weighting strategy may also be easily extended to stream-based ALR, and classification algorithms.", "date": "2026-04-02", "categories": []}
{"id": "2605.01706", "title": "Exploring Entropy-based Active Learning for Fair Brain Segmentation", "abstract": "Active learning (AL) has emerged as a crucial strategy for reducing the prohibitive costs associated with medical image segmentation. However, standard uncertainty-based AL methods typically focus on maximizing performance metrics, ignoring performance disparities or fairness across groups with sensitive attributes. While fair active learning has been explored in classification tasks, its intersection with medical image segmentation remains unaddressed. In this work, we introduced a fairness-aware active learning framework with a Weighted Entropy selection strategy that modulates uncertainty based on current group-specific performance estimates on the labeled set. To decouple true epistemic uncertainty from anatomical volume variances, we further utilized a masked, scaled entropy restricted to the region of interest. The framework was evaluated on synthetic T1-weighted brain MRIs with controlled left caudate bias in both strong and weak bias settings. A 3D U-Net was trained to segment the left caudate under several AL strategies, starting from both demographically balanced and strongly imbalanced initial labeled sets. Experiments demonstrated that our method markedly reduces performance disparities between groups compared to random sampling and standard uncertainty sampling. By prioritizing poorly segmented subgroups during the AL cycles, our method consistently achieved the highest equity-scaled performance and reduced the disparity metric by 75% (strong bias) and 86% (weak bias) relative to standard entropy at the final budget. Overall, this work is among the first studies on fair AL for medical image segmentation, offering an efficient strategy to train more equitable models in resource-constrained environments.", "date": "2026-05-03", "categories": []}
{"id": "2605.02458", "title": "Active multiple matrix completion with adaptive confidence sets", "abstract": "In this work, we formulate a new multi-task active learning setting in which the learner's goal is to solve multiple matrix completion problems simultaneously. At each round, the learner can choose from which matrix it receives a sample from an entry drawn uniformly at random. Our main practical motivation is market segmentation, where the matrices represent different regions with different preferences of the customers. The challenge in this setting is that each of the matrices can be of a different size and also of a different rank which is unknown. We provide and analyze a new algorithm, MAlocate that is able to adapt to the unknown ranks of the different matrices. We then give a lower-bound showing that our strategy is minimax-optimal and demonstrate its performance with synthetic experiments.", "date": "2026-05-04", "categories": []}
{"id": "2510.23557", "title": "Minimizing Human Intervention in Online Classification", "abstract": "Training or fine-tuning large language model (LLM)-based systems often requires costly human feedback, yet there is limited understanding of how to minimize such intervention while maintaining strong error guarantees. We study this problem for LLM-based classification systems in an active learning framework: an agent sequentially labels $d$-dimensional query embeddings drawn i.i.d. from an unknown distribution by either calling a costly expert or guessing with no feedback, with the goal of minimizing regret relative to an oracle with free expert access. When the horizon $T$ is at least exponential in the embedding dimension $d$, the geometry of the class regions can be learned. In this regime, we propose the Conservative Hull-based Classifier (CHC), which maintains convex hulls of expert-labeled queries and calls the expert when a query lands outside all known hulls. CHC attains $\\mathcal{O}(\\log^d T)$ regret in $T$ and is minimax optimal for $d=1$. Otherwise, the geometry cannot be reliably learned in general. We show that for queries drawn from a subgaussian mixture and $T \\le e^d$, a Center-based Classifier (CC) achieves regret proportional to $N\\log{N}$ where $N$ is the number of labels. To bridge these regimes, we introduce the Generalized Hull-based Classifier (GHC), a practical extension of CHC that enables more aggressive guessing via a tunable parameter. Our approach is validated on real-world question-answering datasets using state-of-the-art text embedding models.", "date": "2025-10-27", "categories": []}
{"id": "2605.00480", "title": "Leveraging Vision-Language Models as Weak Annotators in Active Learning", "abstract": "Active learning aims to reduce annotation cost by selectively querying informative samples for supervision under a limited labeling budget. In this work, we investigate how vision-language models (VLMs) can be leveraged to further reduce the reliance on costly human annotation within the active learning paradigm. To this end, we find that the reliability of VLMs varies significantly with label granularity in fine-grained recognition tasks: they perform poorly on fine-grained labels but can provide accurate coarse-grained labels. Leveraging this property, we propose an active learning framework that combines fine-grained human annotations with coarse-grained VLM-generated weak labels through instance-wise label assignment. We further model the systematic noise in VLM-generated labels using a small set of trusted full labels. Experiments on CUB200 and FGVC-Aircraft show that the proposed framework consistently outperforms existing active learning methods under the same annotation budget.", "date": "2026-05-01", "categories": []}
{"id": "2604.22990", "title": "Hard to See, Hard to Label: Generative and Symbolic Acquisition for Subtle Visual Phenomena", "abstract": "Subtle visual anomalies such as hairline cracks, sub-millimeter voids, and low-contrast inclusions are structurally atypical yet visually ambiguous, making them both difficult to annotate and easy to overlook during active learning. Standard acquisition heuristics based on discriminative uncertainty or feature diversity often overselect dominant patterns while underexploring sparse yet important regions of the data space. This failure mode is especially severe in industrial defect inspection, where anomalies may be both low-prevalence and difficult to distinguish from surrounding structure. To resolve this, we propose GSAL, an active learning framework for object detection that combines a diffusion-based difficulty signal with a hierarchical semantic coverage prior. The diffusion component scores images and proposals using reconstruction discrepancy and denoising variability, prioritizing visually atypical or ambiguous examples. However, diffusion alone does not prevent acquisition from repeatedly favoring hard samples within dominant semantic modes. The semantic component therefore organizes candidate samples in a three-level concept graph and promotes coverage of underrepresented semantic regions while providing interpretable acquisition rationales. By balancing visual difficulty with semantic coverage, GSAL improves retrieval of subtle and rare targets that are often missed by uncertainty-only selection. Experiments on a proprietary thin-film defect, Pascal VOC and MS COCO dataset show consistent gains in label efficiency and rare-class retrieval over uncertainty-, diversity-, and hybrid-based baselines", "date": "2026-04-24", "categories": []}
{"id": "2604.23290", "title": "An Analysis of Active Learning Algorithms using Real-World Crowd-sourced Text Annotations", "abstract": "Active learning algorithms automatically identify the most informative samples from large amounts of unlabeled data and tremendously reduce human annotation effort in inducing a machine learning model. In a conventional active learning setup, the labeling oracles are assumed to be infallible, that is, they always provide correct answers (in terms of class labels) to the queried unlabeled instances, which cannot be guaranteed in real-world applications. To this end, a body of research has focused on the development of active learning algorithms in the presence of imperfect / noisy oracles. Existing research on active learning with noisy oracles typically simulate the oracles using machine learning models; however, real-world situations are much more challenging, and using ML models to simulate the annotation patterns may not appropriately capture the nuances of real-world annotation challenges. In this research, we first collect annotations of text samples (from 3 benchmark text classification datasets) from crowd-sourced workers through a crowd-sourcing platform. We then conduct extensive empirical studies of 8 commonly used active learning techniques (in conjunction with deep neural networks) using the obtained annotations. Our analyses sheds light on the performance of these techniques under real-world challenges, where annotators can provide incorrect labels, and can also refuse to provide labels. We hope this research will provide valuable insights that will be useful for the deployment of deep active learning systems in real-world applications. The obtained annotations can be accessed at https://github.com/varuntotakura/al_rcta/.", "date": "2026-04-25", "categories": []}
{"id": "2604.23821", "title": "Accelerating Quantum Materials Characterization: Hybrid Active Learning for Autonomous Spin Wave Spectroscopy", "abstract": "Autonomous neutron spectroscopy must solve three distinct tasks: detection (where is the signal?), inference (which Hamiltonian governs it?), and refinement (what are the parameters?). No single controller solves all three equally well. We present TAS-AI, a hybrid agnostic-to-physics-informed framework for autonomous triple-axis spin-wave spectroscopy that separates these tasks explicitly. In blind reconstruction benchmarks, model-agnostic methods such as random sampling, coarse grids, and Gaussian-process mappers reach a global error threshold more reliably and with fewer measurements than physics-informed planning, supporting the claim that discovery and inference are distinct tasks requiring distinct controllers. Once signal structure is localized, the physics-informed stage performs in-loop Hamiltonian discrimination and parameter refinement: in a controlled square-lattice test between nearest-neighbor-only and J1-J2 Hamiltonians, TAS-AI reaches a decisive AIC-derived evidence ratio (>100) in fewer than 10 measurements, while motion-aware scheduling cuts wall-clock time by 32% at a fixed measurement budget. We also identify a failure mode of posterior-weighted design, algorithmic myopia, in which the planner over-refines the current leading model while under-sampling low-intensity falsification probes. A constrained falsification channel sharply reduces time spent committed to the wrong model and accelerates correct model selection without modifying the Bayesian inference engine. In controlled two-model ablations, both a deterministic top-two max-disagreement rule and an LLM-based audit committee achieve this gain under identical constraints. We demonstrate the full workflow in silico using a high-fidelity digital twin and provide an open-source Python implementation.", "date": "2026-04-26", "categories": []}
{"id": "2510.03247", "title": "Towards Multimodal Active Learning: Efficient Learning with Limited Paired Data", "abstract": "Active learning (AL) is a principled strategy to reduce annotation cost in data-hungry deep learning. However, existing AL algorithms focus almost exclusively on unimodal data, overlooking the substantial annotation burden in multimodal learning. We introduce the first framework for multimodal active learning with unaligned data, where the learner must actively acquire cross-modal alignments rather than labels on pre-aligned pairs. This setting captures the practical bottleneck in modern multimodal pipelines, where unimodal features are easy to obtain but high-quality alignment is costly. We develop a new algorithm that combines uncertainty and diversity principles in a modality-aware design, achieves linear-time acquisition, and applies seamlessly to both pool-based and streaming-based settings. Extensive experiments on benchmark datasets demonstrate that our approach consistently reduces multimodal annotation cost while preserving performance; for instance, on the ColorSwap dataset it cuts annotation requirements by up to 40% without loss in accuracy.", "date": "2025-09-25", "categories": []}
{"id": "2602.23666", "title": "Active Learning for Planet Habitability Classification under Extreme Class Imbalance", "abstract": "The increasing size and heterogeneity of exoplanet catalogs have made systematic habitability assessment challenging, particularly given the extreme scarcity of potentially habitable planets and the evolving nature of their labels. In this study, we explore the use of pool-based active learning to improve the efficiency of habitability classification under realistic observational constraints. We construct a unified dataset from the Habitable World Catalog and the NASA Exoplanet Archive and formulate habitability assessment as a binary classification problem. A supervised baseline based on gradient-boosted decision trees is established and optimized for recall in order to prioritize the identification of rare potentially habitable planets. This model is then embedded within an active learning framework, where uncertainty-based margin sampling is compared against random querying across multiple runs and labeling budgets. We find that active learning substantially reduces the number of labeled instances required to approach supervised performance, demonstrating clear gains in label efficiency. To connect these results to a practical astronomical use case, we aggregate predictions from independently trained active-learning models into an ensemble and use the resulting mean probabilities and uncertainties to rank planets originally labeled as non-habitable. This procedure identifies a single robust candidate for further study, illustrating how active learning can support conservative, uncertainty-aware prioritization of follow-up targets rather than speculative reclassification. Our results indicate that active learning provides a principled framework for guiding habitability studies in data regimes characterized by label imbalance, incomplete information, and limited observational resources.", "date": "2026-02-27", "categories": []}
{"id": "2604.20083", "title": "Energy-Based Open-Set Active Learning for Object Classification", "abstract": "Active learning (AL) has emerged as a crucial methodology for minimizing labeling costs in deep learning by selecting the most valuable samples from a pool of unlabeled data for annotation. Traditional AL operates under a closed-set assumption, where all classes in the dataset are known and consistent. However, real-world scenarios often present open-set conditions in which unlabeled data contains both known and unknown classes. In such environments, standard AL techniques struggle. They can mistakenly query samples from unknown categories, leading to inefficient use of annotation budgets. In this paper, we propose a novel dual-stage energy-based framework for open-set AL. Our method employs two specialized energy-based models (EBMs). The first, an energy-based known/unknown separator, filters out samples likely to belong to unknown classes. The second, an energy-based sample scorer, assesses the informativeness of the filtered known samples. Using the energy landscape, our models distinguish between data points from known and unknown classes in the unlabeled pool by assigning lower energy to known samples and higher energy to unknown samples, ensuring that only samples from classes of interest are selected for labeling. By integrating these components, our approach ensures efficient and targeted sample selection, maximizing learning impact in each iteration. Experiments on 2D (CIFAR-10, CIFAR-100, TinyImageNet) and 3D (ModelNet40) object classification benchmarks demonstrates that our framework outperforms existing approaches, achieving superior annotation efficiency and classification performance in open-set environments.", "date": "2026-04-22", "categories": []}
{"id": "2604.18862", "title": "Human-Machine Co-Boosted Bug Report Identification with Mutualistic Neural Active Learning", "abstract": "Bug reports, encompassing a wide range of bug types, are crucial for maintaining software quality. However, the increasing complexity and volume of bug reports pose a significant challenge in sole manual identification and assignment to the appropriate teams for resolution, as dealing with all the reports is time-consuming and resource-intensive. In this paper, we introduce a cross-project framework, dubbed Mutualistic Neural Active Learning (MNAL), designed for automated and more effective identification of bug reports from GitHub repositories boosted by human-machine collaboration. MNAL utilizes a neural language model that learns and generalizes reports across different projects, coupled with active learning to form neural active learning. A distinctive feature of MNAL is the purposely crafted mutualistic relation between the machine learners (neural language model) and human labelers (developers) when enriching the knowledge learned. That is, the most informative human-labeled reports and their corresponding pseudo-labeled ones are used to update the model while those reports that need to be labeled by developers are more readable and identifiable, thereby enhancing the human-machine teaming therein. We evaluate MNAL using a large scale dataset against the SOTA approaches, baselines, and different variants. The results indicate that MNAL achieves up to 95.8% and 196.0% effort reduction in terms of readability and identifiability during human labeling, respectively, while resulting in a better performance in bug report identification. Additionally, our MNAL is model-agnostic since it is capable of improving the model performance with various underlying neural language models. To further verify the efficacy of our approach, we conducted a qualitative case study involving 10 human participants, who rate MNAL as being more effective while saving more time and monetary resources.", "date": "2026-04-20", "categories": []}
{"id": "2604.17906", "title": "Bayesian Active Learning with Gaussian Processes Guided by LLM Relevance Scoring for Dense Passage Retrieval", "abstract": "While Large Language Models (LLMs) exhibit exceptional zero-shot relevance modeling, their high computational cost necessitates framing passage retrieval as a budget-constrained global optimization problem. Existing approaches passively rely on first-stage dense retrievers, which leads to two limitations: (1) failing to retrieve relevant passages in semantically distinct clusters, and (2) failing to propagate relevance signals to the broader corpus. To address these limitations, we propose Bayesian Active Learning with Gaussian Processes guided by LLM relevance scoring (BAGEL), a novel framework that propagates sparse LLM relevance signals across the embedding space to guide global exploration. BAGEL models the multimodal relevance distribution across the entire embedding space with a query-specific Gaussian Process (GP) based on LLM relevance scores. Subsequently, it iteratively selects passages for scoring by strategically balancing the exploitation of high-confidence regions with the exploration of uncertain areas. Extensive experiments across four benchmark datasets and two LLM backbones demonstrate that BAGEL effectively explores and captures complex relevance distributions and outperforms LLM reranking methods under the same LLM budget on all four datasets.", "date": "2026-04-20", "categories": []}
{"id": "2604.18569", "title": "Revisiting Active Sequential Prediction-Powered Mean Estimation", "abstract": "In this work, we revisit the problem of active sequential prediction-powered mean estimation, where at each round one must decide the query probability of the ground-truth label upon observing the covariates of a sample. Furthermore, if the label is not queried, the prediction from a machine learning model is used instead. Prior work proposed an elegant scheme that determines the query probability by combining an uncertainty-based suggestion with a constant probability that encodes a soft constraint on the query probability. We explored different values of the mixing parameter and observed an intriguing empirical pattern: the smallest confidence width tends to occur when the weight on the constant probability is close to one, thereby reducing the influence of the uncertainty-based component. Motivated by this observation, we develop a non-asymptotic analysis of the estimator and establish a data-dependent bound on its confidence interval. Our analysis further suggests that when a no-regret learning approach is used to determine the query probability and control this bound, the query probability converges to the constraint of the max value of the query probability when it is chosen obliviously to the current covariates. We also conduct simulations that corroborate these theoretical findings.", "date": "2026-04-20", "categories": []}
{"id": "2510.25892", "title": "Topology-Aware Active Learning on Graphs", "abstract": "We propose a graph-topological approach to active learning that directly targets the core challenge of exploration versus exploitation under scarce label budgets. To guide exploration, we introduce a coreset construction algorithm based on Balanced Forman Curvature (BFC), which selects representative initial labels that reflect the graph's cluster structure. This method includes a data-driven stopping criterion that signals when the graph has been sufficiently explored. We further use BFC to dynamically trigger the shift from exploration to exploitation within active learning routines, replacing hand-tuned heuristics. To improve exploitation, we introduce a localized graph rewiring strategy that efficiently incorporates multiscale information around labeled nodes, enhancing label propagation while preserving sparsity. Experiments on benchmark classification tasks show that our methods consistently outperform existing graph-based semi-supervised baselines at low label rates.", "date": "2025-10-29", "categories": []}
{"id": "2604.12303", "title": "Labeled TrustSet Guided: Batch Active Learning with Reinforcement Learning", "abstract": "Batch active learning (BAL) is a crucial technique for reducing labeling costs and improving data efficiency in training large-scale deep learning models. Traditional BAL methods often rely on metrics like Mahalanobis Distance to balance uncertainty and diversity when selecting data for annotation. However, these methods predominantly focus on the distribution of unlabeled data and fail to leverage feedback from labeled data or the model's performance. To address these limitations, we introduce TrustSet, a novel approach that selects the most informative data from the labeled dataset, ensuring a balanced class distribution to mitigate the long-tail problem. Unlike CoreSet, which focuses on maintaining the overall data distribution, TrustSet optimizes the model's performance by pruning redundant data and using label information to refine the selection process. To extend the benefits of TrustSet to the unlabeled pool, we propose a reinforcement learning (RL)-based sampling policy that approximates the selection of high-quality TrustSet candidates from the unlabeled data. Combining TrustSet and RL, we introduce the Batch Reinforcement Active Learning with TrustSet (BRAL-T) framework. BRAL-T achieves state-of-the-art results across 10 image classification benchmarks and 2 active fine-tuning tasks, demonstrating its effectiveness and efficiency in various domains.", "date": "2026-04-14", "categories": []}
{"id": "2501.19227", "title": "Integrating Semi-Supervised and Active Learning for Semantic Segmentation", "abstract": "In this paper, we propose a novel active learning approach integrated with an improved semi-supervised learning framework to reduce the cost of manual annotation and enhance model performance. Our proposed approach effectively leverages both the labelled data selected through active learning and the unlabelled data excluded from the selection process. The proposed active learning approach pinpoints areas where the pseudo-labels are likely to be inaccurate. Then, an automatic and efficient pseudo-label auto-refinement (PLAR) module is proposed to correct pixels with potentially erroneous pseudo-labels by comparing their feature representations with those of labelled regions. This approach operates without increasing the labelling budget and is based on the cluster assumption, which states that pixels belonging to the same class should exhibit similar representations in feature space. Furthermore, manual labelling is only applied to the most difficult and uncertain areas in unlabelled data, where insufficient information prevents the PLAR module from making a decision. We evaluated the proposed hybrid semi-supervised active learning framework on two benchmark datasets, one from natural and the other from remote sensing imagery domains. In both cases, it outperformed state-of-the-art methods in the semantic segmentation task.", "date": "2025-01-31", "categories": []}
{"id": "2503.02983", "title": "BLADE: Bayesian Langevin Active Discovery with Replica Exchange for Identification of Complex Systems", "abstract": "Traditional methods for system discovery frequently struggle with efficient data usage and uncertainty quantification. Identifying the governing equations of complex dynamical systems from data presents a significant challenge in scientific discovery, especially when high-quality measurements are scarce and expensive to obtain. To overcome these limitations, we propose Bayesian Langevin Active Discovery with Replica Exchange for Identification of Complex Systems (BLADE), a novel Bayesian framework that combines replica-exchange stochastic gradient Langevin Monte Carlo with active learning. By balancing gradient-driven exploration and exploitation in coefficient space, BLADE provides probabilistic parameter estimation and principled uncertainty quantification. Faced with data scarcity, the probabilistic foundation of BLADE further facilitates the integration of active learning through a hybrid acquisition strategy that combines predictive uncertainty with space-filling design, enabling efficient selection of informative samples. Across benchmark systems, BLADE reduces measurement requirements by roughly 60% for Lotka-Volterra and 40% for Burgers' equation relative to random sampling, demonstrating substantial data-efficiency gains. These results highlight BLADE as a general uncertainty-aware framework for discovering interpretable dynamical systems, particularly valuable when high-fidelity data acquisition is prohibitively expensive.", "date": "2025-03-04", "categories": []}
{"id": "2604.08965", "title": "Dynamic Class-Aware Active Learning for Unbiased Satellite Image Segmentation", "abstract": "Semantic segmentation of satellite imagery plays a vital role in land cover mapping and environmental monitoring. However, annotating large-scale, high-resolution satellite datasets is costly and time consuming, especially when covering vast geographic regions. Instead of randomly labeling data or exhaustively annotating entire datasets, Active Learning (AL) offers an efficient alternative by intelligently selecting the most informative samples for annotation with the help of Human-in-the-loop (HITL), thereby reducing labeling costs while maintaining high model performance. AL is particularly beneficial for large-scale or resource-constrained satellite applications, as it enables high segmentation accuracy with significantly fewer labeled samples. Despite these advantages, standard AL strategies typically rely on global uncertainty or diversity measures and lack the adaptability to target underperforming or rare classes as training progresses, leading to bias in the system. To overcome these limitations, we propose a novel adaptive acquisition function, Dynamic Class-Aware Uncertainty based Active learning (DCAU-AL) that prioritizes sample selection based on real-time class-wise performance gaps, thereby overcoming class-imbalance issue. The proposed DCAU-AL mechanism continuously tracks the performance of the segmentation per class and dynamically adjusts the sampling weights to focus on poorly performing or underrepresented classes throughout the active learning process. Extensive experiments on the OpenEarth land cover dataset show that DCAU-AL significantly outperforms existing AL methods, especially under severe class imbalance, delivering superior per-class IoU and improved annotation efficiency.", "date": "2026-04-10", "categories": []}
{"id": "2604.08977", "title": "Testing the Assumptions of Active Learning for Translation Tasks with Few Samples", "abstract": "Active learning (AL) is a training paradigm for selecting unlabeled samples for annotation to improve model performance on a test set, which is useful when only a limited number of samples can be annotated. These algorithms often work by optimizing for the informativeness and diversity of the training data to be annotated. Recent work found that AL strategies fail to outperform random sampling on various language generation tasks when using 100-500 samples. To understand AL's poor performance when only using few samples, we investigate whether the core assumptions underlying AL strategies hold. We find that neither the informativeness nor diversity of the training data, which AL strategies optimize for, are correlated with test set performance. Instead, factors like the ordering of the training samples and interactions with pre-training data have a larger impact on performance. This suggests that future AL methods must take these factors into account in order to work with very few samples.", "date": "2026-04-10", "categories": []}
{"id": "2403.03208", "title": "Active Statistical Inference", "abstract": "Inspired by the concept of active learning, we propose active inference$\\unicode{x2013}$a methodology for statistical inference with machine-learning-assisted data collection. Assuming a budget on the number of labels that can be collected, the methodology uses a machine learning model to identify which data points would be most beneficial to label, thus effectively utilizing the budget. It operates on a simple yet powerful intuition: prioritize the collection of labels for data points where the model exhibits uncertainty, and rely on the model's predictions where it is confident. Active inference constructs provably valid confidence intervals and hypothesis tests while leveraging any black-box machine learning model and handling any data distribution. The key point is that it achieves the same level of accuracy with far fewer samples than existing baselines relying on non-adaptively-collected data. This means that for the same number of collected samples, active inference enables smaller confidence intervals and more powerful p-values. We evaluate active inference on datasets from public opinion research, census analysis, and proteomics.", "date": "2024-03-05", "categories": []}
{"id": "2603.24480", "title": "Positive-First Most Ambiguous: A Simple Active Learning Criterion for Interactive Retrieval of Rare Categories", "abstract": "Real-world fine-grained visual retrieval often requires discovering a rare concept from large unlabeled collections with minimal supervision. This is especially critical in biodiversity monitoring, ecological studies, and long-tailed visual domains, where the target may represent only a tiny fraction of the data, creating highly imbalanced binary problems. Interactive retrieval with relevance feedback offers a practical solution: starting from a small query, the system selects candidates for binary user annotation and iteratively refines a lightweight classifier. While Active Learning (AL) is commonly used to guide selection, conventional AL assumes symmetric class priors and large annotation budgets, limiting effectiveness in imbalanced, low-budget, low-latency settings. We introduce Positive-First Most Ambiguous (PF-MA), a simple yet effective AL criterion that explicitly addresses the class imbalance asymmetry: it prioritizes near-boundary samples while favoring likely positives, enabling rapid discovery of subtle visual categories while maintaining informativeness. Unlike standard methods that oversample negatives, PF-MA consistently returns small batches with a high proportion of relevant samples, improving early retrieval and user satisfaction. To capture retrieval diversity, we also propose a class coverage metric that measures how well selected positives span the visual variability of the target class. Experiments on long-tailed datasets, including fine-grained botanical data, demonstrate that PF-MA consistently outperforms strong baselines in both coverage and classifier performance, across varying class sizes and descriptors. Our results highlight that aligning AL with the asymmetric and user-centric objectives of interactive fine-grained retrieval enables simple yet powerful solutions for retrieving rare and visually subtle categories in realistic human-in-the-loop settings.", "date": "2026-03-25", "categories": []}
{"id": "2307.02719", "title": "Understanding Uncertainty Sampling via Equivalent Loss", "abstract": "Uncertainty sampling is a prevalent active learning algorithm that queries sequentially the annotations of data samples which the current prediction model is uncertain about. However, the usage of uncertainty sampling has been largely heuristic: There is no consensus on the proper definition of ``uncertainty'' for a specific task under a specific loss, nor a theoretical guarantee that prescribes a standard protocol to implement the algorithm. In this work, we systematically examine uncertainty sampling algorithms in the binary classification problem via a notion of equivalent loss which depends on the used uncertainty measure and the original loss function, and establish that an uncertainty sampling algorithm is optimizing against such an equivalent loss. The perspective verifies the properness of existing uncertainty measures from two aspects: surrogate property and loss convexity. When the convexity is preserved, we give a sample complexity result for the equivalent loss, and later translate it into a binary loss guarantee via the surrogate link function. We prove the asymptotic superiority of the uncertainty sampling against the passive learning via this approach under mild conditions. We also discuss some potential extensions, including pool-based setting and potential generalization to the multi-class classification as well as the regression problems.", "date": "2023-07-06", "categories": []}
{"id": "2604.05227", "title": "Active Measurement of Two-Point Correlations", "abstract": "Two-point correlation functions (2PCF) are widely used to characterize how points cluster in space. In this work, we study the problem of measuring the 2PCF over a large set of points, restricted to a subset satisfying a property of interest. An example comes from astronomy, where scientists measure the 2PCF of star clusters, which make up only a tiny subset of possible sources within a galaxy. This task typically requires careful labeling of sources to construct catalogs, which is time-consuming. We present a human-in-the-loop framework for efficient estimation of 2PCF of target sources. By leveraging a pre-trained classifier to guide sampling, our approach adaptively selects the most informative points for human annotation. After each annotation, it produces unbiased estimates of pair counts across multiple distance bins simultaneously. Compared to simple Monte Carlo approaches, our method achieves substantially lower variance while significantly reducing annotation effort. We introduce a novel unbiased estimator, sampling strategy, and confidence interval construction that together enable scalable and statistically grounded measurement of two-point correlations in astronomy datasets.", "date": "2026-04-06", "categories": []}
{"id": "2604.05259", "title": "Coverage Optimization for Camera View Selection", "abstract": "What makes a good viewpoint? The quality of the data used to learn 3D reconstructions is crucial for enabling efficient and accurate scene modeling. We study the active view selection problem and develop a principled analysis that yields a simple and interpretable criterion for selecting informative camera poses. Our key insight is that informative views can be obtained by minimizing a tractable approximation of the Fisher Information Gain, which reduces to favoring viewpoints that cover geometry that has been insufficiently observed by past cameras. This leads to a lightweight coverage-based view selection metric that avoids expensive transmittance estimation and is robust to noise and training dynamics. We call this metric COVER (Camera Optimization for View Exploration and Reconstruction). We integrate our method into the Nerfstudio framework and evaluate it on real datasets within fixed and embodied data acquisition scenarios. Across multiple datasets and radiance-field baselines, our method consistently improves reconstruction quality compared to state-of-the-art active view selection methods. Additional visualizations and our Nerfstudio package can be found at https://chengine.github.io/nbv_gym/.", "date": "2026-04-06", "categories": []}
{"id": "2501.15458", "title": "Amortized Safe Active Learning for Real-Time Data Acquisition: Pretrained Neural Policies From Simulated Nonparametric Functions", "abstract": "Safe active learning (AL) is a sequential scheme for learning unknown systems while respecting safety constraints during data acquisition. Existing methods often rely on Gaussian processes (GPs) to model the task and safety constraints, requiring repeated GP updates and constrained acquisition optimization--incurring significant computations which are challenging for real-time decision-making. We propose amortized AL for regression and amortized safe AL, replacing expensive online computations with a pretrained neural policy. Inspired by recent advances in amortized Bayesian experimental design, we leverage GPs as pretraining simulators. We train our policy prior to the AL deployment on simulated nonparametric functions, using Fourier feature-based GP sampling and a differentiable acquisition objective that is safety-aware in the safe AL setting. At deployment, our policy selects informative and (if desired) safe queries via a single forward pass, eliminating GP inference and acquisition optimization. This leads to magnitudes of speed improvements while preserving learning quality. Our framework is modular and, without the safety component, yields fast unconstrained AL for time-sensitive tasks.", "date": "2025-01-26", "categories": []}
{"id": "2604.02766", "title": "Random Is Hard to Beat: Active Selection in online DPO with Modern LLMs", "abstract": "Modern LLMs inherit strong priors from web-scale pretraining, which can limit the headroom of post-training data-selection strategies. While Active Preference Learning (APL) seeks to optimize query efficiency in online Direct Preference Optimization (DPO), the inherent richness of on-policy candidate pools often renders simple Random sampling a surprisingly formidable baseline. We evaluate uncertainty-based APL against Random across harmlessness, helpfulness, and instruction-following settings, utilizing both reward models and LLM-as-a-judge proxies. We find that APL yields negligible improvements in proxy win-rates compared to Random. Crucially, we observe a dissociation where win-rate improves even as general capability -- measured by standard benchmarks -- degrades. APL fails to mitigate this capability collapse or reduce variance significantly better than random sampling. Our findings suggest that in the regime of strong pre-trained priors, the computational overhead of active selection is difficult to justify against the ``cheap diversity'' provided by simple random samples. Our code is available at https://github.com/BootsofLagrangian/random-vs-apl.", "date": "2026-04-03", "categories": []}
{"id": "2504.16624", "title": "A Detailed Account of Compositional Automata Learning through Alphabet Refinement", "abstract": "Active automata learning infers automaton models of systems from behavioral observations, a technique successfully applied to a wide range of domains. Compositional approaches have recently emerged to address scalability to concurrent systems. We take a significant step beyond available results, including those by the authors, and develop a general technique for compositional learning of a synchronizing parallel system with an unknown decomposition. Our approach automatically refines the global alphabet into component alphabets while learning the component models. We develop a theoretical treatment of distributions of alphabets, i.e., sets of possibly overlapping component alphabets, characterize counter-examples that reveal inconsistencies with global observations, and show how to systematically update the distribution to restore consistency. We extend $L^{\\star}$ to handle partial and potentially spurious information arising when learning components from global observations only. We establish correctness and termination of the full algorithm. We provide an implementation, called CoalA, using the state-of-the-art active learning library LearnLib. Our experiments on more than 630 subject systems show that CoalA delivers up to five orders of magnitude fewer membership queries than monolithic learning, and achieves better scalability in equivalence queries on systems with significant concurrency.", "date": "2025-04-23", "categories": []}
{"id": "2603.29135", "title": "Quality-Controlled Active Learning via Gaussian Processes for Robust Structure-Property Learning in Autonomous Microscopy", "abstract": "Autonomous experimental systems are increasingly used in materials research to accelerate scientific discovery, but their performance is often limited by low-quality, noisy data. This issue is especially problematic in data-intensive structure-property learning tasks such as Image-to-Spectrum (Im2Spec) and Spectrum-to-Image (Spec2Im) translations, where standard active learning strategies can mistakenly prioritize poor-quality measurements. We introduce a gated active learning framework that combines curiosity-driven sampling with a physics-informed quality control filter based on the Simple Harmonic Oscillator model fits, allowing the system to automatically exclude low-fidelity data during acquisition. Evaluations on a pre-acquired dataset of band-excitation piezoresponse spectroscopy (BEPS) data from PbTiO3 thin films with spatially localized noise show that the proposed method outperforms random sampling, standard active learning, and multitask learning strategies. The gated approach enhances both Im2Spec and Spec2Im by handling noise during training and acquisition, leading to more reliable forward and inverse predictions. In contrast, standard active learners often misinterpret noise as uncertainty and end up acquiring bad samples that hurt performance. Given its promising applicability, we further deployed the framework in real-time experiments on BiFeO3 thin films, demonstrating its effectiveness in real autonomous microscopy experiments. Overall, this work supports a shift toward hybrid autonomy in self-driving labs, where physics-informed quality assessment and active decision-making work hand-in-hand for more reliable discovery.", "date": "2026-03-31", "categories": []}
{"id": "2603.29677", "title": "Mind the Gap: A Framework for Assessing Pitfalls in Multimodal Active Learning", "abstract": "Multimodal learning enables neural networks to integrate information from heterogeneous sources, but active learning in this setting faces distinct challenges. These include missing modalities, differences in modality difficulty, and varying interaction structures. These are issues absent in the unimodal case. While the behavior of active learning strategies in unimodal settings is well characterized, their behavior under such multimodal conditions remains poorly understood. We introduce a new framework for benchmarking multimodal active learning that isolates these pitfalls using synthetic datasets, allowing systematic evaluation without confounding noise. Using this framework, we compare unimodal and multimodal query strategies and validate our findings on two real-world datasets. Our results show that models consistently develop imbalanced representations, relying primarily on one modality while neglecting others. Existing query methods do not mitigate this effect, and multimodal strategies do not consistently outperform unimodal ones. These findings highlight limitations of current active learning methods and underline the need for modality-aware query strategies that explicitly address these pitfalls. Code and benchmark resources will be made publicly available.", "date": "2026-03-31", "categories": []}
{"id": "2501.00277", "title": "Efficient Human-in-the-Loop Active Learning: A Novel Framework for Data Labeling in AI Systems", "abstract": "Modern AI algorithms require labeled data. In real world, majority of data are unlabeled. Labeling the data are costly. this is particularly true for some areas requiring special skills, such as reading radiology images by physicians. To most efficiently use expert's time for the data labeling, one promising approach is human-in-the-loop active learning algorithm. In this work, we propose a novel active learning framework with significant potential for application in modern AI systems. Unlike the traditional active learning methods, which only focus on determining which data point should be labeled, our framework also introduces an innovative perspective on incorporating different query scheme. We propose a model to integrate the information from different types of queries. Based on this model, our active learning frame can automatically determine how the next question is queried. We further developed a data driven exploration and exploitation framework into our active learning method. This method can be embedded in numerous active learning algorithms. Through simulations on five real-world datasets, including a highly complex real image task, our proposed active learning framework exhibits higher accuracy and lower loss compared to other methods.", "date": "2024-12-31", "categories": []}
{"id": "2603.27385", "title": "Active In-Context Learning for Tabular Foundation Models", "abstract": "Active learning (AL) reduces labeling cost by querying informative samples, but in tabular settings its cold-start gains are often limited because uncertainty estimates are unreliable when models are trained on very few labels. Tabular foundation models such as TabPFN provide calibrated probabilistic predictions via in-context learning (ICL), i.e., without task-specific weight updates, enabling an AL regime in which the labeled context - rather than parameters - is iteratively optimized. We formalize Tabular Active In-Context Learning (Tab-AICL) and instantiate it with four acquisition rules: uncertainty (TabPFN-Margin), diversity (TabPFN-Coreset), an uncertainty-diversity hybrid (TabPFN-Hybrid), and a scalable two-stage method (TabPFN-Proxy-Hybrid) that shortlists candidates using a lightweight linear proxy before TabPFN-based selection. Across 20 classification benchmarks, Tab-AICL improves cold-start sample efficiency over retrained gradient-boosting baselines (CatBoost-Margin and XGBoost-Margin), measured by normalized AULC up to 100 labeled samples.", "date": "2026-03-28", "categories": []}
{"id": "2511.22344", "title": "Cleaning the Pool: Progressive Filtering of Unlabeled Pools in Deep Active Learning", "abstract": "Existing active learning (AL) strategies capture fundamentally different notions of data value, e.g., uncertainty or representativeness. Consequently, the effectiveness of strategies can vary substantially across datasets, models, and even AL cycles. Committing to a single strategy risks suboptimal performance, as no single strategy dominates throughout the entire AL process. We introduce REFINE, an ensemble AL method that combines multiple strategies without knowing in advance which will perform best. In each AL cycle, REFINE operates in two stages: (1) Progressive filtering iteratively refines the unlabeled pool by considering an ensemble of AL strategies, retaining promising candidates capturing different notions of value. (2) Coverage-based selection then chooses a final batch from this refined pool, ensuring all previously identified notions of value are accounted for. Extensive experiments across 6 classification datasets and 3 foundation models show that REFINE consistently outperforms individual strategies and existing ensemble methods. Notably, progressive filtering serves as a powerful preprocessing step that improves the performance of any individual AL strategy applied to the refined pool, which we demonstrate on an audio spectrogram classification use case. Finally, the ensemble of REFINE can be easily extended with upcoming state-of-the-art AL strategies.", "date": "2025-11-27", "categories": []}
{"id": "2603.23159", "title": "Conformal Cross-Modal Active Learning", "abstract": "Foundation models for vision have transformed visual recognition with powerful pretrained representations and strong zero-shot capabilities, yet their potential for data-efficient learning remains largely untapped. Active Learning (AL) aims to minimize annotation costs by strategically selecting the most informative samples for labeling, but existing methods largely overlook the rich multimodal knowledge embedded in modern vision-language models (VLMs). We introduce Conformal Cross-Modal Acquisition (CCMA), a novel AL framework that bridges vision and language modalities through a teacher-student architecture. CCMA employs a pretrained VLM as a teacher to provide semantically grounded uncertainty estimates, conformally calibrated to guide sample selection for a vision-only student model. By integrating multimodal conformal scoring with diversity-aware selection strategies, CCMA achieves superior data efficiency across multiple benchmarks. Our approach consistently outperforms state-of-the-art AL baselines, demonstrating clear advantages over methods relying solely on uncertainty or diversity metrics.", "date": "2026-03-24", "categories": []}
{"id": "2603.25107", "title": "Label What Matters: Modality-Balanced and Difficulty-Aware Multimodal Active Learning", "abstract": "Multimodal learning integrates complementary information from different modalities such as image, text, and audio to improve model performance, but its success relies on large-scale labeled data, which is costly to obtain. Active learning (AL) mitigates this challenge by selectively annotating informative samples. In multimodal settings, many approaches implicitly assume that modality importance is stable across rounds and keep selection rules fixed at the fusion stage, which leaves them insensitive to the dynamic nature of multimodal learning, where the relative value of modalities and the difficulty of instances shift as training proceeds. To address this issue, we propose RL-MBA, a reinforcement-learning framework for modality-balanced, difficulty-aware multimodal active learning. RL-MBA models sample selection as a Markov Decision Process, where the policy adapts to modality contributions, uncertainty, and diversity, and the reward encourages accuracy gains and balance. Two key components drive this adaptability: (1) Adaptive Modality Contribution Balancing (AMCB), which dynamically adjusts modality weights via reinforcement feedback, and (2) Evidential Fusion for DifficultyAware Policy Adjustment (EFDA), which estimates sample difficulty via uncertainty-based evidential fusion to prioritize informative samples. Experiments on Food101, KineticsSound, and VGGSound demonstrate that RL-MBA consistently outperforms strong baselines, improving both classification accuracy and modality fairness under limited labeling budgets.", "date": "2026-03-26", "categories": []}
{"id": "2602.07440", "title": "Active Learning Using Aggregated Acquisition Functions: Accuracy and Sustainability Analysis", "abstract": "Active learning (AL) is a machine learning (ML) approach that strategically selects the most informative samples for annotation during training, aiming to minimize annotation costs. This strategy not only reduces labeling expenses but also results in energy savings during neural network training, thereby enhancing both data and energy efficiency. In this paper, we implement and evaluate various state-of-the-art acquisition functions, analyzing their accuracy and computational costs, while discussing the advantages and disadvantages of each method. Our findings reveal that representativity-based acquisition functions effectively explore the dataset but do not prioritize boundary decisions, whereas uncertainty-based acquisition functions focus on refining boundary decisions already identified by the neural network. This trade-off is known as the exploration-exploitation dilemma. To address this dilemma, we introduce six aggregation structures: series, parallel, hybrid, adaptive feedback, random exploration, and annealing exploration. Our aggregated acquisition functions alleviate common AL pathologies such as batch mode inefficiency and the cold start problem. Additionally, we focus on balancing accuracy and energy consumption, contributing to the development of more sustainable, energy-aware artificial intelligence (AI). We evaluate our proposed structures on various models and datasets. Our results demonstrate the potential of these structures to reduce computational costs while maintaining or even improving accuracy. Innovative aggregation approaches, such as alternating between acquisition functions such as BALD and BADGE, have shown robust results. Sequentially running functions like $K$-Centers followed by BALD has achieved the same performance goals with up to 12\\% fewer samples, while reducing the acquisition cost by almost half.", "date": "2026-02-07", "categories": []}
{"id": "2603.12567", "title": "Foundation-Model Surrogates Enable Data-Efficient Active Learning for Materials Discovery", "abstract": "Active learning (AL) has emerged as a powerful paradigm for accelerating materials discovery by iteratively steering experiments toward promising candidates, reducing the number of costly synthesis-and-characterization cycles needed to identify optimal materials. However, current AL relies predominantly on Gaussian Process (GP) and Random Forest (RF) surrogates, which suffer from complementary limitations: GP underfits complex composition-property landscapes due to rigid kernel assumptions, while RF produces unreliable heuristic uncertainty estimates in small-data regimes. This small-data challenge is pervasive in materials science, making reliable surrogate modeling extremely difficult with models trained from scratch on each new dataset. Here we propose In-Context Active Learning (ICAL), which addresses this bottleneck by replacing conventional surrogates with TabPFN, a transformer-based foundation model (FM) pre-trained on millions of synthetic regression tasks to meta-learn a universal prior over tabular data, upon which TabPFN performs principled Bayesian inference in a single forward pass without dataset-specific retraining, delivering strong small-data regression performance and well-calibrated predictive uncertainty (required for effective AL). We benchmark ICAL against GP and RF across 10 materials datasets and TabPFN wins on 8 out of 10 datasets, achieving a mean saving of 52% in extra evaluations relative to GP and 29.77% relative to RF. Cross-validation analysis confirms that TabPFN's advantage stems from superior uncertainty calibration, achieving the lowest Negative Log-Likelihood and Area Under the Sparsification Error curve among all surrogates. These results demonstrate that pre-trained FMs can serve as effective surrogates for active learning, enabling data-efficient discovery across diverse materials systems and small-data experimental sciences.", "date": "2026-03-13", "categories": []}
{"id": "2603.22750", "title": "REALITrees: Rashomon Ensemble Active Learning for Interpretable Trees", "abstract": "Active learning reduces labeling costs by selecting samples that maximize information gain. A dominant framework, Query-by-Committee (QBC), typically relies on perturbation-based diversity by inducing model disagreement through random feature subsetting or data blinding. While this approximates one notion of epistemic uncertainty, it sacrifices direct characterization of the plausible hypothesis space. We propose the complementary approach: Rashomon Ensembled Active Learning (REAL) which constructs a committee by exhaustively enumerating the Rashomon Set of all near-optimal models. To address functional redundancy within this set, we adopt a PAC-Bayesian framework using a Gibbs posterior to weight committee members by their empirical risk. Leveraging recent algorithmic advances, we exactly enumerate this set for the class of sparse decision trees. Across synthetic and established active learning baselines, REAL outperforms randomized ensembles, particularly in moderately noisy environments where it strategically leverages expanded model multiplicity to achieve faster convergence.", "date": "2026-03-24", "categories": []}
{"id": "2412.09686", "title": "The Cost of Replicability in Active Learning", "abstract": "Active learning aims to reduce the number of labeled data points required by machine learning algorithms by selectively querying labels from initially unlabeled data. Ensuring replicability, where an algorithm produces consistent outcomes across different runs, is essential for the reliability of machine learning models but often increases sample complexity. This paper investigates the cost of replicability in active learning using two classical disagreement-based methods: the CAL and A^2 algorithms. Leveraging randomized thresholding techniques, we propose two replicable active learning algorithms: one for realizable learning of finite hypothesis classes and another for the agnostic setting. Our theoretical analysis shows that while enforcing replicability increases label complexity, CAL and A^2 still achieve substantial label savings under this constraint. These findings provide insights into balancing efficiency and stability in active learning.", "date": "2024-12-12", "categories": []}
{"id": "2603.21840", "title": "Select, Label, Evaluate: Active Testing in NLP", "abstract": "Human annotation cost and time remain significant bottlenecks in Natural Language Processing (NLP), with test data annotation being particularly expensive due to the stringent requirement for low-error and high-quality labels necessary for reliable model evaluation. Traditional approaches require annotating entire test sets, leading to substantial resource requirements. Active Testing is a framework that selects the most informative test samples for annotation. Given a labeling budget, it aims to choose the subset that best estimates model performance while minimizing cost and human effort. In this work, we formalize Active Testing in NLP and we conduct an extensive benchmarking of existing approaches across 18 datasets and 4 embedding strategies spanning 4 different NLP tasks. The experiments show annotation reductions of up to 95%, with performance estimation accuracy difference from the full test set within 1%. Our analysis reveals variations in method effectiveness across different data characteristics and task types, with no single approach emerging as universally superior. Lastly, to address the limitation of requiring a predefined annotation budget in existing sample selection strategies, we introduce an adaptive stopping criterion that automatically determines the optimal number of samples.", "date": "2026-03-23", "categories": []}
{"id": "2603.22160", "title": "Data Curation for Machine Learning Interatomic Potentials by Determinantal Point Processes", "abstract": "The development of machine learning interatomic potentials faces a critical computational bottleneck with the generation and labeling of useful training datasets. We present a novel application of determinantal point processes (DPPs) to the task of selecting informative subsets of atomic configurations to label with reference energies and forces from costly quantum mechanical methods. Through experiments with hafnium oxide data, we show that DPPs are competitive with existing approaches to constructing compact but diverse training sets by utilizing kernels of molecular descriptors, leading to improved accuracy and robustness in machine learning representations of molecular systems. Our work identifies promising directions to employ DPPs for unsupervised training data curation with heterogeneous or multimodal data, or in online active learning schemes for iterative data augmentation during molecular dynamics simulation.", "date": "2026-03-23", "categories": []}
{"id": "2603.19264", "title": "Generative Active Testing: Efficient LLM Evaluation via Proxy Task Adaptation", "abstract": "With the widespread adoption of pre-trained Large Language Models (LLM), there exists a high demand for task-specific test sets to benchmark their performance in domains such as healthcare and biomedicine. However, the cost of labeling test samples while developing new benchmarks poses a significant challenge, especially when expert annotators are required. Existing frameworks for active sample selection offer limited support for generative Question Answering tasks, where option dynamics can affect model decision boundaries. In this paper, we present Generative Active Testing (GAT), an uncertainty-aware acquisition framework leveraging LLMs as surrogates for informing the sample selection process. Using a novel Statement Adaptation Module, we modify generative tasks into a pseudo-classification format, enabling the capture of sample-level uncertainties across unlabeled candidates. Our zero-shot acquisition functions reduce estimation error by ~40% compared to traditional sampling baselines, offering a scalable solution for cost-effective model benchmarking.", "date": "2026-02-26", "categories": []}
{"id": "2603.11370", "title": "Relaxed Efficient Acquisition of Context and Temporal Features", "abstract": "In many biomedical applications, measurements are not freely available at inference time: each laboratory test, imaging modality, or assessment incurs financial cost, time burden, or patient risk. Longitudinal active feature acquisition (LAFA) seeks to optimize predictive performance under such constraints by adaptively selecting measurements over time, yet the problem remains inherently challenging due to temporally coupled decisions (missed early measurements cannot be revisited, and acquisition choices influence all downstream predictions). Moreover, real-world clinical workflows typically begin with an initial onboarding phase, during which relatively stable contextual descriptors (e.g., demographics or baseline characteristics) are collected once and subsequently condition longitudinal decision-making. Despite its practical importance, the efficient selection of onboarding context has not been studied jointly with temporally adaptive acquisition. We therefore propose REACT (Relaxed Efficient Acquisition of Context and Temporal features), an end-to-end differentiable framework that simultaneously optimizes (i) selection of onboarding contextual descriptors and (ii) adaptive feature--time acquisition plans for longitudinal measurements under cost constraints. REACT employs a Gumbel--Sigmoid relaxation with straight-through estimation to enable gradient-based optimization over discrete acquisition masks, allowing direct backpropagation from prediction loss and acquisition cost. Across real-world longitudinal health and behavioral datasets, REACT achieves improved predictive performance at lower acquisition costs compared to existing longitudinal acquisition baselines, demonstrating the benefit of modeling onboarding and temporally coupled acquisition within a unified optimization framework.", "date": "2026-03-11", "categories": []}
{"id": "2603.16715", "title": "Novelty-Driven Target-Space Discovery in Automated Electron and Scanning Probe Microscopy", "abstract": "Modern automated microscopy faces a fundamental discovery challenge: in many systems, the most important scientific information does not reside in the immediately visible image features, but in the target space of sequentially acquired spectra or functional responses, making it essential to develop strategies that can actively search for new behaviors rather than simply optimize known objectives. Here, we developed a deep-kernel-learning BEACON framework that is explicitly designed to guide discovery in the target space by learning structure-property relationships during the experiment and using that evolving model to seek diverse response regimes. We first established the method through demonstration workflows built on pre-acquired ground-truth datasets, which enabled direct benchmarking against classical acquisition strategies and allowed us to define a set of monitoring functions for comparing exploration quality, target-space coverage, and surrogate-model behavior in a transparent and reproducible manner. This benchmarking framework provides a practical basis for evaluating discovery-driven algorithms, not just optimization performance. We then operationalized and deployed the workflow on STEM, showing that the approach can transition from offline validation to real experimental implementation. To support adoption and extension by the broader community, the associated notebooks are available, allowing users to reproduce the workflows, test the benchmarks, and adapt the method to their own instruments and datasets.", "date": "2026-03-17", "categories": []}
{"id": "2603.15336", "title": "Active Seriation: Efficient Ordering Recovery with Statistical Guarantees", "abstract": "Active seriation aims at recovering an unknown ordering of $n$ items by adaptively querying pairwise similarities. The observations are noisy measurements of entries of an underlying $n$ x $n$ permuted Robinson matrix, whose permutation encodes the latent ordering. The framework allows the algorithm to start with partial information on the latent ordering, including seriation from scratch as a special case. We propose an active seriation algorithm that provably recovers the latent ordering with high probability. Under a uniform separation condition on the similarity matrix, optimal performance guarantees are established, both in terms of the probability of error and the number of observations required for successful recovery.", "date": "2026-03-16", "categories": []}
{"id": "2603.13109", "title": "BoSS: A Best-of-Strategies Selector as an Oracle for Deep Active Learning", "abstract": "Active learning (AL) aims to reduce annotation costs while maximizing model performance by iteratively selecting valuable instances. While foundation models have made it easier to identify these instances, existing selection strategies still lack robustness across different models, annotation budgets, and datasets. To highlight the potential weaknesses of existing AL strategies and provide a reference point for research, we explore oracle strategies, i.e., strategies that approximate the optimal selection by accessing ground-truth information unavailable in practical AL scenarios. Current oracle strategies, however, fail to scale effectively to large datasets and complex deep neural networks. To tackle these limitations, we introduce the Best-of-Strategy Selector (BoSS), a scalable oracle strategy designed for large-scale AL scenarios. BoSS constructs a set of candidate batches through an ensemble of selection strategies and then selects the batch yielding the highest performance gain. As an ensemble of selection strategies, BoSS can be easily extended with new state-of-the-art strategies as they emerge, ensuring it remains a reliable oracle strategy in the future. Our evaluation demonstrates that i) BoSS outperforms existing oracle strategies, ii) state-of-the-art AL strategies still fall noticeably short of oracle performance, especially in large-scale datasets with many classes, and iii) one possible solution to counteract the inconsistent performance of AL strategies might be to employ an ensemble-based approach for the selection.", "date": "2026-03-13", "categories": []}
{"id": "2210.06112", "title": "Efficient Bayesian Updates for Deep Active Learning via Laplace Approximations", "abstract": "Deep active learning (AL) selects batches of instances for annotation to avoid retraining deep neural networks (DNNs) after each new label. Employing a naive top-$b$ selection can result in a batch of redundant (similar) instances. To address this, various AL strategies employ clustering techniques that ensure diversity within a batch. We approach this issue by substituting the costly retraining with an efficient Bayesian update. Our proposed update represents a second-order optimization step using the Gaussian posterior from a last-layer Laplace approximation. Thereby, we achieve low computational complexity by computing the inverse Hessian in closed form. We demonstrate that in typical AL settings, our update closely approximates retraining while being considerably faster. Leveraging our update, we introduce a new framework for batch selection through sequential construction, updating the DNN after each label acquisition. Furthermore, we incorporate our update into a look-ahead selection strategy as a feasible upper baseline approximating optimal batch selection. Our results highlight the potential of efficient updates to advance deep AL research.", "date": "2022-10-12", "categories": []}
{"id": "2404.08981", "title": "Fast Fishing: Approximating BAIT for Efficient and Scalable Deep Active Image Classification", "abstract": "Deep active learning (AL) seeks to minimize the annotation costs for training deep neural networks. BAIT, a recently proposed AL strategy based on the Fisher Information, has demonstrated impressive performance across various datasets. However, BAIT's high computational and memory requirements hinder its applicability on large-scale classification tasks, resulting in current research neglecting BAIT in their evaluation. This paper introduces two methods to enhance BAIT's computational efficiency and scalability. Notably, we significantly reduce its time complexity by approximating the Fisher Information. In particular, we adapt the original formulation by i) taking the expectation over the most probable classes, and ii) constructing a binary classification task, leading to an alternative likelihood for gradient computations. Consequently, this allows the efficient use of BAIT on large-scale datasets, including ImageNet. Our unified and comprehensive evaluation across a variety of datasets demonstrates that our approximations achieve strong performance with considerably reduced time complexity. Furthermore, we provide an extensive open-source toolbox that implements recent state-of-the-art AL strategies, available at https://github.com/dhuseljic/dal-toolbox.", "date": "2024-04-13", "categories": []}
{"id": "2602.18867", "title": "Similarity-as-Evidence: Calibrating Overconfident VLMs for Interpretable and Label-Efficient Medical Active Learning", "abstract": "Active Learning (AL) reduces annotation costs in medical imaging by selecting only the most informative samples for labeling, but suffers from cold-start when labeled data are scarce. Vision-Language Models (VLMs) address the cold-start problem via zero-shot predictions, yet their temperature-scaled softmax outputs treat text-image similarities as deterministic scores while ignoring inherent uncertainty, leading to overconfidence. This overconfidence misleads sample selection, wasting annotation budgets on uninformative cases. To overcome these limitations, the Similarity-as-Evidence (SaE) framework calibrates text-image similarities by introducing a Similarity Evidence Head (SEH), which reinterprets the similarity vector as evidence and parameterizes a Dirichlet distribution over labels. In contrast to a standard softmax that enforces confident predictions even under weak signals, the Dirichlet formulation explicitly quantifies lack of evidence (vacuity) and conflicting evidence (dissonance), thereby mitigating overconfidence caused by rigid softmax normalization. Building on this, SaE employs a dual-factor acquisition strategy: high-vacuity samples (e.g., rare diseases) are prioritized in early rounds to ensure coverage, while high-dissonance samples (e.g., ambiguous diagnoses) are prioritized later to refine boundaries, providing clinically interpretable selection rationales. Experiments on ten public medical imaging datasets with a 20% label budget show that SaE attains state-of-the-art macro-averaged accuracy of 82.57%. On the representative BTMRI dataset, SaE also achieves superior calibration, with a negative log-likelihood (NLL) of 0.425.", "date": "2026-02-21", "categories": []}
{"id": "2603.10341", "title": "Federated Active Learning Under Extreme Non-IID and Global Class Imbalance", "abstract": "Federated active learning (FAL) seeks to reduce annotation cost under privacy constraints, yet its effectiveness degrades in realistic settings with severe global class imbalance and highly heterogeneous clients. We conduct a systematic study of query-model selection in FAL and uncover a central insight: the model that achieves more class-balanced sampling, especially for minority classes, consistently leads to better final performance. Moreover, global-model querying is beneficial only when the global distribution is highly imbalanced and client data are relatively homogeneous; otherwise, the local model is preferable. Based on these findings, we propose FairFAL, an adaptive class-fair FAL framework. FairFAL (1) infers global imbalance and local-global divergence via lightweight prediction discrepancy, enabling adaptive selection between global and local query models; (2) performs prototype-guided pseudo-labeling using global features to promote class-aware querying; and (3) applies a two-stage uncertainty-diversity balanced sampling strategy with k-center refinement. Experiments on five benchmarks show that FairFAL consistently outperforms state-of-the-art approaches under challenging long-tailed and non-IID settings. The code is available at https://github.com/chenchenzong/FairFAL.", "date": "2026-03-11", "categories": []}
{"id": "2603.10435", "title": "Adaptive Active Learning for Regression via Reinforcement Learning", "abstract": "Active learning for regression reduces labeling costs by selecting the most informative samples. Improved Greedy Sampling is a prominent method that balances feature-space diversity and output-space uncertainty using a static, multiplicative rule. We propose Weighted improved Greedy Sampling (WiGS), which replaces this framework with a dynamic, additive criterion. We formulate weight selection as a reinforcement learning problem, enabling an agent to adapt the exploration-investigation balance throughout learning. Experiments on 18 benchmark datasets and a synthetic environment show WiGS outperforms iGS and other baseline methods in both accuracy and labeling efficiency, particularly in domains with irregular data density where the baseline's multiplicative rule ignores high-error samples in dense regions.", "date": "2026-03-11", "categories": []}
{"id": "2411.16722", "title": "Active Prompt Learning with Vision-Language Model Priors", "abstract": "Vision-language models (VLMs) have demonstrated remarkable zero-shot performance across various classification tasks. Nonetheless, their reliance on hand-crafted text prompts for each task hinders efficient adaptation to new tasks. While prompt learning offers a promising solution, most studies focus on maximizing the utilization of given few-shot labeled datasets, often overlooking the potential of careful data selection strategies, which enable higher accuracy with fewer labeled data. This motivates us to study a budget-efficient active prompt learning framework. Specifically, we introduce a class-guided clustering that leverages the pre-trained image and text encoders of VLMs, thereby enabling our cluster-balanced acquisition function from the initial round of active learning. Furthermore, considering the substantial class-wise variance in confidence exhibited by VLMs, we propose a budget-saving selective querying based on adaptive class-wise thresholds. Extensive experiments in active learning scenarios across seven datasets demonstrate that our method outperforms existing baselines.", "date": "2024-11-23", "categories": []}
{"id": "2509.25376", "title": "Cold-Start Active Correlation Clustering", "abstract": "We study active correlation clustering where pairwise similarities are not provided upfront and must be queried in a cost-efficient manner through active learning. Specifically, we focus on the cold-start scenario, where no true initial pairwise similarities are available for active learning. To address this challenge, we propose a coverage-aware method that encourages diversity early in the process. We demonstrate the effectiveness of our approach through several synthetic and real-world experiments.", "date": "2025-09-29", "categories": []}
{"id": "2603.07163", "title": "PromptGate Client Adaptive Vision Language Gating for Open Set Federated Active Learning", "abstract": "Deploying medical AI across resource-constrained institutions demands data-efficient learning pipelines that respect patient privacy. Federated Learning (FL) enables collaborative medical AI without centralising data, yet real-world clinical pools are inherently open-set, containing out-of-distribution (OOD) noise such as imaging artifacts and wrong modalities. Standard Active Learning (AL) query strategies mistake this noise for informative samples, wasting scarce annotation budgets. We propose PromptGate, a dynamic VLM-gated framework for Open-Set Federated AL that purifies unlabeled pools before querying. PromptGate introduces a federated Class-Specific Context Optimization: lightweight, learnable prompt vectors that adapt a frozen BiomedCLIP backbone to local clinical domains and aggregate globally via FedAvg -- without sharing patient data. As new annotations arrive, prompts progressively sharpen the ID/OOD boundary, turning the VLM into a dynamic gatekeeper that is strategy-agnostic: a plug-and-play pre-selection module enhancing any downstream AL strategy. Experiments on distributed dermatology and breast imaging benchmarks show that while static VLM prompting degrades to 50% ID purity, PromptGate maintains $>$95% purity with 98% OOD recall.", "date": "2026-03-07", "categories": []}
{"id": "2603.07898", "title": "Revisiting Unknowns: Towards Effective and Efficient Open-Set Active Learning", "abstract": "Open-set active learning (OSAL) aims to identify informative samples for annotation when unlabeled data may contain previously unseen classes-a common challenge in safety-critical and open-world scenarios. Existing approaches typically rely on separately trained open-set detectors, introducing substantial training overhead and overlooking the supervisory value of labeled unknowns for improving known-class learning. In this paper, we propose E$^2$OAL (Effective and Efficient Open-set Active Learning), a unified and detector-free framework that fully exploits labeled unknowns for both stronger supervision and more reliable querying. E$^2$OAL first uncovers the latent class structure of unknowns through label-guided clustering in a frozen contrastively pre-trained feature space, optimized by a structure-aware F1-product objective. To leverage labeled unknowns, it employs a Dirichlet-calibrated auxiliary head that jointly models known and unknown categories, improving both confidence calibration and known-class discrimination. Building on this, a logit-margin purity score estimates the likelihood of known classes to construct a high-purity candidate pool, while an OSAL-specific informativeness metric prioritizes partially ambiguous yet reliable samples. These components together form a flexible two-stage query strategy with adaptive precision control and minimal hyperparameter sensitivity. Extensive experiments across multiple OSAL benchmarks demonstrate that E$^2$OAL consistently surpasses state-of-the-art methods in accuracy, efficiency, and query precision, highlighting its effectiveness and practicality for real-world applications. The code is available at github.com/chenchenzong/E2OAL.", "date": "2026-03-09", "categories": []}
{"id": "2406.02566", "title": "Combining X-Vectors and Bayesian Batch Active Learning: Two-Stage Active Learning Pipeline for Speech Recognition", "abstract": "This paper introduces a novel two-stage active learning (AL) pipeline for automatic speech recognition (ASR), combining unsupervised and supervised AL methods. The first stage utilizes unsupervised AL by using x-vectors clustering for diverse sample selection from unlabeled speech data, thus establishing a robust initial dataset for the subsequent supervised AL. The second stage incorporates a supervised AL strategy, with a batch AL method specifically developed for ASR, aimed at selecting diverse and informative batches of samples. Here, sample diversity is also achieved using x-vectors clustering, while the most informative samples are identified using a Bayesian AL method tailored for ASR with an adaptation of Monte Carlo dropout to approximate Bayesian inference. This approach enables precise uncertainty estimation, thereby enhancing ASR model training with significantly reduced data requirements. Our method has shown superior performance compared to competing methods on homogeneous, heterogeneous, and OOD test sets, demonstrating that strategic sample selection and innovative Bayesian modeling can substantially optimize both labeling effort and data utilization in deep learning-based ASR applications.", "date": "2024-05-03", "categories": []}
{"id": "2602.13308", "title": "Learning to Select Like Humans: Explainable Active Learning for Medical Imaging", "abstract": "Medical image analysis requires substantial labeled data for model training, yet expert annotation is expensive and time-consuming. Active learning (AL) addresses this challenge by strategically selecting the most informative samples for the annotation purpose, but traditional methods solely rely on predictive uncertainty while ignoring whether models learn from clinically meaningful features a critical requirement for clinical deployment. We propose an explainability-guided active learning framework that integrates spatial attention alignment into a sample acquisition process. Our approach advocates for a dual-criterion selection strategy combining: (i) classification uncertainty to identify informative examples, and (ii) attention misalignment with radiologist-defined regions-of-interest (ROIs) to target samples where the model focuses on incorrect features. By measuring misalignment between Grad-CAM attention maps and expert annotations using Dice similarity, our acquisition function judiciously identifies samples that enhance both predictive performance and spatial interpretability. We evaluate the framework using three expert-annotated medical imaging datasets, namely, BraTS (MRI brain tumors), VinDr-CXR (chest X-rays), and SIIM-COVID-19 (chest X-rays). Using only 570 strategically selected samples, our explainability-guided approach consistently outperforms random sampling across all the datasets, achieving 77.22% accuracy on BraTS, 52.37% on VinDr-CXR, and 52.66% on SIIM-COVID. Grad-CAM visualizations confirm that the models trained by our dual-criterion selection focus on diagnostically relevant regions, demonstrating that incorporating explanation guidance into sample acquisition yields superior data efficiency while maintaining clinical interpretability.", "date": "2026-02-10", "categories": []}
{"id": "2603.01171", "title": "PARWiS: Winner determination under shoestring budgets using active pairwise comparisons", "abstract": "Determining a winner among a set of items using active pairwise comparisons under a limited budget is a challenging problem in preference-based learning. The goal of this study is to implement and evaluate the PARWiS algorithm, which shows spectral ranking and disruptive pair selection to identify the best item under shoestring budgets. This work have extended the PARWiS with a contextual variant (Contextual PARWiS) and a reinforcement learning-based variant (RL PARWiS), comparing them against baselines, including Double Thompson Sampling and a random selection strategy. This evaluation spans synthetic and real-world datasets (Jester and MovieLens), using budgets of 40, 60, and 80 comparisons for 20 items. The performance is measured through recovery fraction, true rank of reported winner, reported rank of true winner, and cumulative regret, alongside the separation metric \\(\\Delta_{1,2}\\). Results show that PARWiS and RL PARWiS outperform baselines across all datasets, particularly in the Jester dataset with a higher \\(\\Delta_{1,2}\\), while performance gaps narrow in the more challenging MovieLens dataset with a smaller \\(\\Delta_{1,2}\\). Contextual PARWiS shows comparable performance to PARWiS, indicating that contextual features may require further tuning to provide significant benefits.", "date": "2026-03-01", "categories": []}
{"id": "2602.24263", "title": "Active Bipartite Ranking with Smooth Posterior Distributions", "abstract": "In this article, bipartite ranking, a statistical learning problem involved in many applications and widely studied in the passive context, is approached in a much more general \\textit{active setting} than the discrete one previously considered in the literature. While the latter assumes that the conditional distribution is piece wise constant, the framework we develop permits in contrast to deal with continuous conditional distributions, provided that they fulfill a H\\\"older smoothness constraint. We first show that a naive approach based on discretisation at a uniform level, fixed \\textit{a priori} and consisting in applying next the active strategy designed for the discrete setting generally fails. Instead, we propose a novel algorithm, referred to as smooth-rank and designed for the continuous setting, which aims to minimise the distance between the ROC curve of the estimated ranking rule and the optimal one w.r.t. the $\\sup$ norm. We show that, for a fixed confidence level $\\epsilon>0$ and probability $\\delta\\in (0,1)$, smooth-rank is PAC$(\\epsilon,\\delta)$. In addition, we provide a problem dependent upper bound on the expected sampling time of smooth-rank and establish a problem dependent lower bound on the expected sampling time of any PAC$(\\epsilon,\\delta)$ algorithm. Beyond the theoretical analysis carried out, numerical results are presented, providing solid empirical evidence of the performance of the algorithm proposed, which compares favorably with alternative approaches.", "date": "2026-02-27", "categories": []}
{"id": "2602.21674", "title": "Error-awareness Accelerates Active Automata Learning", "abstract": "Active automata learning (AAL) algorithms can learn a behavioral model of a system from interacting with it. The primary challenge remains scaling to larger models, in particular in the presence of many possible inputs to the system. Modern AAL algorithms fail to scale even if, in every state, most inputs lead to errors. In various challenging problems from the literature, these errors are observable, i.e., they emit a known error output. Motivated by these problems, we study learning these systems more efficiently. Further, we consider various degrees of knowledge about which inputs are non-error producing at which state. For each level of knowledge, we provide a matching adaptation of the state-of-the-art AAL algorithm L# to make the most of this domain knowledge. Our empirical evaluation demonstrates that the methods accelerate learning by orders of magnitude with strong but realistic domain knowledge to a single order of magnitude with limited domain knowledge.", "date": "2026-02-25", "categories": []}
{"id": "2602.19578", "title": "Goal-Oriented Influence-Maximizing Data Acquisition for Learning and Optimization", "abstract": "Active data acquisition is central to many learning and optimization tasks in deep neural networks, yet remains challenging because most approaches rely on predictive uncertainty estimates that are difficult to obtain reliably. To this end, we propose Goal-Oriented Influence- Maximizing Data Acquisition (GOIMDA), an active acquisition algorithm that avoids explicit posterior inference while remaining uncertainty-aware through inverse curvature. GOIMDA selects inputs by maximizing their expected influence on a user-specified goal functional, such as test loss, predictive entropy, or the value of an optimizer-recommended design. Leveraging first-order influence functions, we derive a tractable acquisition rule that combines the goal gradient, training-loss curvature, and candidate sensitivity to model parameters. We show theoretically that, for generalized linear models, GOIMDA approximates predictive-entropy minimization up to a correction term accounting for goal alignment and prediction bias, thereby, yielding uncertainty-aware behavior without maintaining a Bayesian posterior. Empirically, across learning tasks (including image and text classification) and optimization tasks (including noisy global optimization benchmarks and neural-network hyperparameter tuning), GOIMDA consistently reaches target performance with substantially fewer labeled samples or function evaluations than uncertainty-based active learning and Gaussian-process Bayesian optimization baselines.", "date": "2026-02-23", "categories": []}
{"id": "2601.20775", "title": "Active Learning for Decision Trees with Provable Guarantees", "abstract": "This paper advances the theoretical understanding of active learning label complexity for decision trees as binary classifiers. We make two main contributions. First, we provide the first analysis of the disagreement coefficient for decision trees-a key parameter governing active learning label complexity. Our analysis holds under two natural assumptions required for achieving polylogarithmic label complexity, (i) each root-to-leaf path queries distinct feature dimensions, and (ii) the input data has a regular, grid-like structure. We show these assumptions are essential, as relaxing them leads to polynomial label complexity. Second, we present the first general active learning algorithm for binary classification that achieves a multiplicative error guarantee, producing a $(1+\\epsilon)$-approximate classifier. By combining these results, we design an active learning algorithm for decision trees that uses only a polylogarithmic number of label queries in the dataset size, under the stated assumptions. Finally, we establish a label complexity lower bound, showing our algorithm's dependence on the error tolerance $\\epsilon$ is close to optimal.", "date": "2026-01-28", "categories": []}
{"id": "2602.15738", "title": "Beyond Labels: Information-Efficient Human-in-the-Loop Learning using Ranking and Selection Queries", "abstract": "Integrating human expertise into machine learning systems often reduces the role of experts to labeling oracles, a paradigm that limits the amount of information exchanged and fails to capture the nuances of human judgment. We address this challenge by developing a human-in-the-loop framework to learn binary classifiers with rich query types, consisting of item ranking and exemplar selection. We first introduce probabilistic human response models for these rich queries motivated by the relationship experimentally observed between the perceived implicit score of an item and its distance to the unknown classifier. Using these models, we then design active learning algorithms that leverage the rich queries to increase the information gained per interaction. We provide theoretical bounds on sample complexity and develop a tractable and computationally efficient variational approximation. Through experiments with simulated annotators derived from crowdsourced word-sentiment and image-aesthetic datasets, we demonstrate significant reductions on sample complexity. We further extend active learning strategies to select queries that maximize information rate, explicitly balancing informational value against annotation cost. This algorithm in the word sentiment classification task reduces learning time by more than 57\\% compared to traditional label-only active learning.", "date": "2026-02-17", "categories": []}
{"id": "2510.14581", "title": "Model-agnostic Selective Labeling with Provable Statistical Guarantees", "abstract": "Obtaining high-quality labels for large datasets is expensive, requiring massive annotations from human experts. While AI models offer a cost-effective alternative by predicting labels, their label quality is compromised by the unavoidable labeling errors. Existing methods mitigate this issue through selective labeling, where AI labels a subset and human labels the remainder. However, these methods lack theoretical guarantees on the quality of AI-assigned labels, often resulting in unacceptably high labeling error within the AI-labeled subset. To address this, we introduce \\textbf{Conformal Labeling}, a novel method to identify instances where AI predictions can be provably trusted. This is achieved by controlling the false discovery rate (FDR), the proportion of incorrect labels within the selected subset. In particular, we construct a conformal $p$-value for each test instance by comparing AI models' predicted confidence to those of calibration instances mislabeled by AI models. Then, we select test instances whose $p$-values are below a data-dependent threshold, certifying AI models' predictions as trustworthy. We provide theoretical guarantees that Conformal Labeling controls the FDR below the nominal level, ensuring that a predefined fraction of AI-assigned labels is correct on average. Extensive experiments demonstrate that our method achieves tight FDR control with high power across various tasks, including image and text labeling, and LLM QA.", "date": "2025-10-16", "categories": []}
{"id": "2602.13359", "title": "The Speed-up Factor: A Quantitative Multi-Iteration Active Learning Performance Metric", "abstract": "Machine learning models excel with abundant annotated data, but annotation is often costly and time-intensive. Active learning (AL) aims to improve the performance-to-annotation ratio by using query methods (QMs) to iteratively select the most informative samples. While AL research focuses mainly on QM development, the evaluation of this iterative process lacks appropriate performance metrics. This work reviews eight years of AL evaluation literature and formally introduces the speed-up factor, a quantitative multi-iteration QM performance metric that indicates the fraction of samples needed to match random sampling performance. Using four datasets from diverse domains and seven QMs of various types, we empirically evaluate the speed-up factor and compare it with state-of-the-art AL performance metrics. The results confirm the assumptions underlying the speed-up factor, demonstrate its accuracy in capturing the described fraction, and reveal its superior stability across iterations.", "date": "2026-02-13", "categories": []}
{"id": "2510.25926", "title": "Active Learning with Task-Driven Representations for Messy Pools", "abstract": "Active learning has the potential to be especially useful for messy, uncurated pools where datapoints vary in relevance to the target task. However, state-of-the-art approaches to this problem currently rely on using fixed, unsupervised representations of the pool, focusing on modifying the acquisition function instead. We show that this model setup can undermine their effectiveness at dealing with messy pools, as such representations can fail to capture important information relevant to the task. To address this, we propose using task-driven representations that are periodically updated during the active learning process using the previously collected labels. We introduce two specific strategies for learning these representations, one based on directly learning semi-supervised representations and the other based on supervised fine-tuning of an initial unsupervised representation. We find that both significantly improve empirical performance over using unsupervised or pretrained representations.", "date": "2025-10-29", "categories": []}
{"id": "2602.12391", "title": "High-dimensional Level Set Estimation with Trust Regions and Double Acquisition Functions", "abstract": "Level set estimation (LSE) classifies whether an unknown function's value exceeds a specified threshold for given inputs, a fundamental problem in many real-world applications. In active learning settings with limited initial data, we aim to iteratively acquire informative points to construct an accurate classifier for this task. In high-dimensional spaces, this becomes challenging where the search volume grows exponentially with increasing dimensionality. We propose TRLSE, an algorithm for high-dimensional LSE, which identifies and refines regions near the threshold boundary with dual acquisition functions operating at both global and local levels. We provide a theoretical analysis of TRLSE's accuracy and show its superior sample efficiency against existing methods through extensive evaluations on multiple synthetic and real-world LSE problems.", "date": "2026-02-12", "categories": []}
{"id": "2601.13458", "title": "Labels or Preferences? Budget-Constrained Learning with Human Judgments over AI-Generated Outputs", "abstract": "The increasing reliance on human preference feedback to judge AI-generated pseudo labels has created a pressing need for principled, budget-conscious data acquisition strategies. We address the crucial question of how to optimally allocate a fixed annotation budget between ground-truth labels and pairwise preferences in AI. Our solution, grounded in semi-parametric inference, casts the budget allocation problem as a monotone missing data framework. Building on this formulation, we introduce Preference-Calibrated Active Learning (PCAL), a novel method that learns the optimal data acquisition strategy and develops a statistically efficient estimator for functionals of the data distribution. Theoretically, we prove the asymptotic optimality of our PCAL estimator and establish a key robustness guarantee that ensures robust performance even with poorly estimated nuisance models. Our flexible framework applies to a general class of problems, by directly optimizing the estimator's variance instead of requiring a closed-form solution. This work provides a principled and statistically efficient approach for budget-constrained learning in modern AI. Simulations and real-data analysis demonstrate the practical benefits and superior performance of our proposed method.", "date": "2026-01-19", "categories": []}
{"id": "2602.11825", "title": "CAAL: Confidence-Aware Active Learning for Heteroscedastic Atmospheric Regression", "abstract": "Quantifying the impacts of air pollution on health and climate relies on key atmospheric particle properties such as toxicity and hygroscopicity. However, these properties typically require complex observational techniques or expensive particle-resolved numerical simulations, limiting the availability of labeled data. We therefore estimate these hard-to-measure particle properties from routinely available observations (e.g., air pollutant concentrations and meteorological conditions). Because routine observations only indirectly reflect particle composition and structure, the mapping from routine observations to particle properties is noisy and input-dependent, yielding a heteroscedastic regression setting. With a limited and costly labeling budget, the central challenge is to select which samples to measure or simulate. While active learning is a natural approach, most acquisition strategies rely on predictive uncertainty. Under heteroscedastic noise, this signal conflates reducible epistemic uncertainty with irreducible aleatoric uncertainty, causing limited budgets to be wasted in noise-dominated regions. To address this challenge, we propose a confidence-aware active learning framework (CAAL) for efficient and robust sample selection in heteroscedastic settings. CAAL consists of two components: a decoupled uncertainty-aware training objective that separately optimises the predictive mean and noise level to stabilise uncertainty estimation, and a confidence-aware acquisition function that dynamically weights epistemic uncertainty using predicted aleatoric uncertainty as a reliability signal. Experiments on particle-resolved numerical simulations and real atmospheric observations show that CAAL consistently outperforms standard AL baselines. The proposed framework provides a practical and general solution for the efficient expansion of high-cost atmospheric particle property databases.", "date": "2026-02-12", "categories": []}
{"id": "2601.07651", "title": "Active Evaluation of General Agents: Problem Definition and Comparison of Baseline Algorithms", "abstract": "As intelligent agents become more generally-capable, i.e. able to master a wide variety of tasks, the complexity and cost of properly evaluating them rises significantly. Tasks that assess specific capabilities of the agents can be correlated and stochastic, requiring many samples for accurate comparisons, leading to added costs. In this paper, we propose a formal definition and a conceptual framework for active evaluation of agents across multiple tasks, which assesses the performance of ranking algorithms as a function of number of evaluation data samples. Rather than curating, filtering, or compressing existing data sets as a preprocessing step, we propose an online framing: on every iteration, the ranking algorithm chooses the task and agents to sample scores from. Then, evaluation algorithms report a ranking of agents on each iteration and their performance is assessed with respect to the ground truth ranking over time. Several baselines are compared under different experimental contexts, with synthetic generated data and simulated online access to real evaluation data from Atari game-playing agents. We find that the classical Elo rating system -- while it suffers from well-known failure modes, in theory -- is a consistently reliable choice for efficient reduction of ranking error in practice. A recently-proposed method, Soft Condorcet Optimization, shows comparable performance to Elo on synthetic data and significantly outperforms Elo on real Atari agent evaluation. When task variation from the ground truth is high, selecting tasks based on proportional representation leads to higher rate of ranking error reduction.", "date": "2026-01-12", "categories": []}
{"id": "2511.20605", "title": "How to Purchase Labels? A Cost-Effective Approach Using Active Learning Markets", "abstract": "We introduce and analyse active learning markets as a way to purchase labels, in situations where analysts aim to acquire additional data to improve model fitting, or to better train models for predictive analytics applications. This comes in contrast to the many proposals that already exist to purchase features and examples. By originally formalising the market clearing as an optimisation problem, we integrate budget constraints and improvement thresholds into the label acquisition process. We focus on a single-buyer-multiple-seller setup and propose the use of two active learning strategies (variance based and query-by-committee based), paired with distinct pricing mechanisms. They are compared to benchmark baselines including random sampling and a greedy knapsack heuristic. The proposed strategies are validated on real-world datasets from two critical application domains: real estate pricing and energy forecasting. Results demonstrate the robustness of our approach, consistently achieving superior performance with fewer labels acquired compared to conventional methods. Our proposal comprises an easy-to-implement practical solution for optimising data acquisition in resource-constrained environments.", "date": "2025-11-25", "categories": []}
{"id": "2507.12412", "title": "NOCTA: Non-Greedy Objective Cost-Tradeoff Acquisition for Longitudinal Data", "abstract": "In many critical domains, features are not freely available at inference time: each measurement may come with a cost of time, money, and risk. Longitudinal prediction further complicates this setting because both features and labels evolve over time, and missing measurements at earlier timepoints may become permanently unavailable. We propose NOCTA, a Non-Greedy Objective Cost-Tradeoff Acquisition framework that sequentially acquires the most informative features at inference time while accounting for both temporal dynamics and acquisition cost. NOCTA is driven by a novel objective, NOCT, which evaluates a candidate set of future feature-time acquisitions by its expected predictive loss together with its acquisition cost. Since NOCT depends on unobserved future trajectories at inference time, we develop two complementary estimators: (i) NOCT-Contrastive, which learns an embedding of partial observations utilizing the induced distribution over future acquisitions, and (ii) NOCT-Amortized, which directly predicts NOCT for candidate plans with a neural network. Experiments on synthetic and real-world medical datasets demonstrate that both NOCTA estimators outperform existing baselines, achieving higher accuracy at lower acquisition costs.", "date": "2025-07-16", "categories": []}
{"id": "2509.13213", "title": "Density-Aware Farthest Point Sampling", "abstract": "We focus on training machine learning regression models in scenarios where the availability of labeled training data is limited due to computational constraints or high labeling costs. Thus, selecting suitable training sets from unlabeled data is essential for balancing performance and efficiency. For the selection of the training data, we focus on passive and model-agnostic sampling methods that only consider the data feature representations. We derive an upper bound for the expected prediction error of Lipschitz continuous regression models that linearly depends on the weighted fill distance of the training set: a quantity we can estimate simply by considering the data features. We introduce ''Density-Aware Farthest Point Sampling'' (DA-FPS), a novel sampling method. We prove that DA-FPS provides approximate minimizers for a data-driven estimation of the weighted fill distance, thereby aiming at minimizing our derived bound. We conduct experiments using two regression models across three datasets. The results demonstrate that DA-FPS significantly reduces the mean absolute prediction error compared to other sampling strategies.", "date": "2025-09-16", "categories": []}
{"id": "2602.06818", "title": "Wild Guesses and Mild Guesses in Active Concept Learning", "abstract": "Human concept learning is typically active: learners choose which instances to query or test in order to reduce uncertainty about an underlying rule or category. Active concept learning must balance informativeness of queries against the stability of the learner that generates and scores hypotheses. We study this trade-off in a neuro-symbolic Bayesian learner whose hypotheses are executable programs proposed by a large language model (LLM) and reweighted by Bayesian updating. We compare a Rational Active Learner that selects queries to maximize approximate expected information gain (EIG) and the human-like Positive Test Strategy (PTS) that queries instances predicted to be positive under the current best hypothesis. Across concept-learning tasks in the classic Number Game, EIG is effective when falsification is necessary (e.g., compound or exception-laden rules), but underperforms on simple concepts. We trace this failure to a support mismatch between the EIG policy and the LLM proposal distribution: highly diagnostic boundary queries drive the posterior toward regions where the generator produces invalid or overly specific programs, yielding a support-mismatch trap in the particle approximation. PTS is information-suboptimal but tends to maintain proposal validity by selecting \"safe\" queries, leading to faster convergence on simple rules. Our results suggest that \"confirmation bias\" may not be a cognitive error, but rather a rational adaptation for maintaining tractable inference in the sparse, open-ended hypothesis spaces characteristic of human thought.", "date": "2026-02-06", "categories": []}
{"id": "2506.22186", "title": "Thompson Sampling-Based Learning and Control for Unknown Dynamic Systems", "abstract": "Thompson sampling (TS) is a Bayesian randomized exploration strategy that samples options (e.g., system parameters or control laws) from the current posterior and then applies the selected option that is optimal for a task, thereby balancing exploration and exploitation; this makes TS effective for active learning-based controller design. However, TS relies on finite parametric representations, which limits its applicability to more general spaces, which are more commonly encountered in control system design. To address this issue, this work proposes a parameterization method for control law learning using reproducing kernel Hilbert spaces and designs a data-driven active learning control approach. Specifically, the proposed method treats the control law as an element in a function space, allowing the design of control laws without imposing restrictions on the system structure or the form of the controller. A TS framework is proposed in this work to reduce control costs through online exploration and exploitation, and the convergence guarantees are further provided for the learning process. Theoretical analysis shows that the proposed method learns the relationship between control laws and closed-loop performance metrics at an exponential rate, and the upper bound of control regret is also derived. Furthermore, the closed-loop stability of the proposed learning framework is analyzed. Numerical experiments on controlling unknown nonlinear systems validate the effectiveness of the proposed method.", "date": "2025-06-27", "categories": []}
{"id": "2602.05087", "title": "Autodiscover: A reinforcement learning recommendation system for the cold-start imbalance challenge in active learning, powered by graph-aware thompson sampling", "abstract": "Systematic literature reviews (SLRs) are fundamental to evidence-based research, but manual screening is an increasing bottleneck as scientific output grows. Screening features low prevalence of relevant studies and scarce, costly expert decisions. Traditional active learning (AL) systems help, yet typically rely on fixed query strategies for selecting the next unlabeled documents. These static strategies do not adapt over time and ignore the relational structure of scientific literature networks. This thesis introduces AutoDiscover, a framework that reframes AL as an online decision-making problem driven by an adaptive agent. Literature is modeled as a heterogeneous graph capturing relationships among documents, authors, and metadata. A Heterogeneous Graph Attention Network (HAN) learns node representations, which a Discounted Thompson Sampling (DTS) agent uses to dynamically manage a portfolio of query strategies. With real-time human-in-the-loop labels, the agent balances exploration and exploitation under non-stationary review dynamics, where strategy utility changes over time. On the 26-dataset SYNERGY benchmark, AutoDiscover achieves higher screening efficiency than static AL baselines. Crucially, the agent mitigates cold start by bootstrapping discovery from minimal initial labels where static approaches fail. We also introduce TS-Insight, an open-source visual analytics dashboard to interpret, verify, and diagnose the agent's decisions. Together, these contributions accelerate SLR screening under scarce expert labels and low prevalence of relevant studies.", "date": "2026-02-04", "categories": []}
{"id": "2602.05333", "title": "Pool-based Active Learning as Noisy Lossy Compression: Characterizing Label Complexity via Finite Blocklength Analysis", "abstract": "This paper proposes an information-theoretic framework for analyzing the theoretical limits of pool-based active learning (AL), in which a subset of instances is selectively labeled. The proposed framework reformulates pool-based AL as a noisy lossy compression problem by mapping pool observations to noisy symbol observations, data selection to compression, and learning to decoding. This correspondence enables a unified information-theoretic analysis of data selection and learning in pool-based AL. Applying finite blocklength analysis of noisy lossy compression, we derive information-theoretic lower bounds on label complexity and generalization error that serve as theoretical limits for a given learning algorithm under its associated optimal data selection strategy. Specifically, our bounds include terms that reflect overfitting induced by the learning algorithm and the discrepancy between its inductive bias and the target task, and are closely related to established information-theoretic bounds and stability theory, which have not been previously applied to the analysis of pool-based AL. These properties yield a new theoretical perspective on pool-based AL.", "date": "2026-02-05", "categories": []}
{"id": "2602.05690", "title": "Almost Asymptotically Optimal Active Clustering Through Pairwise Observations", "abstract": "We propose a new analysis framework for clustering $M$ items into an unknown number of $K$ distinct groups using noisy and actively collected responses. At each time step, an agent is allowed to query pairs of items and observe bandit binary feedback. If the pair of items belongs to the same (resp.\\ different) cluster, the observed feedback is $1$ with probability $p>1/2$ (resp.\\ $q<1/2$). Leveraging the ubiquitous change-of-measure technique, we establish a fundamental lower bound on the expected number of queries needed to achieve a desired confidence in the clustering accuracy, formulated as a sup-inf optimization problem. Building on this theoretical foundation, we design an asymptotically optimal algorithm in which the stopping criterion involves an empirical version of the inner infimum -- the Generalized Likelihood Ratio (GLR) statistic -- being compared to a threshold. We develop a computationally feasible variant of the GLR statistic and show that its performance gap to the lower bound can be accurately empirically estimated and remains within a constant multiple of the lower bound.", "date": "2026-02-05", "categories": []}
{"id": "2508.18380", "title": "Information Templates: A New Paradigm for Intelligent Active Feature Acquisition", "abstract": "Active feature acquisition (AFA) is an instance-adaptive paradigm in which, at inference time, a policy sequentially chooses which features to acquire (at a cost) before predicting. Existing approaches either train reinforcement learning policies, which deal with a difficult MDP, or greedy policies that cannot account for the joint informativeness of features or require knowledge about the underlying data distribution. To overcome this, we propose Template-based AFA (TAFA), a non-greedy framework that learns a small library of feature templates -- sets of features that are jointly informative -- and uses this library of templates to guide the next feature acquisitions. Through identifying feature templates, the proposed framework not only significantly reduces the action space considered by the policy but also alleviates the need to estimate the underlying data distribution. Extensive experiments on synthetic and real-world datasets show that TAFA outperforms the existing state-of-the-art baselines while achieving lower overall acquisition cost and computation.", "date": "2025-08-25", "categories": []}
{"id": "2602.04340", "title": "Explicit Uncertainty Modeling for Active CLIP Adaptation with Dual Prompt Tuning", "abstract": "Pre-trained vision-language models such as CLIP exhibit strong transferability, yet adapting them to downstream image classification tasks under limited annotation budgets remains challenging. In active learning settings, the model must select the most informative samples for annotation from a large pool of unlabeled data. Existing approaches typically estimate uncertainty via entropy-based criteria or representation clustering, without explicitly modeling uncertainty from the model perspective. In this work, we propose a robust uncertainty modeling framework for active CLIP adaptation based on dual-prompt tuning. We introduce two learnable prompts in the textual branch of CLIP. The positive prompt enhances the discriminability of task-specific textual embeddings corresponding to light-weight tuned visual embeddings, improving classification reliability. Meanwhile, the negative prompt is trained in an reversed manner to explicitly model the probability that the predicted label is correct, providing a principled uncertainty signal for guiding active sample selection. Extensive experiments across different fine-tuning paradigms demonstrate that our method consistently outperforms existing active learning methods under the same annotation budget.", "date": "2026-02-04", "categories": []}
{"id": "2602.02550", "title": "HyPAC: Cost-Efficient LLMs-Human Hybrid Annotation with PAC Error Guarantees", "abstract": "Data annotation often involves multiple sources with different cost-quality trade-offs, such as fast large language models (LLMs), slow reasoning models, and human experts. In this work, we study the problem of routing inputs to the most cost-efficient annotation source while controlling the labeling error on test instances. We propose \\textbf{HyPAC}, a method that adaptively labels inputs to the most cost-efficient annotation source while providing distribution-free guarantees on annotation error. HyPAC calibrates two decision thresholds using importance sampling and upper confidence bounds, partitioning inputs into three regions based on uncertainty and routing each to the appropriate annotation source. We prove that HyPAC achieves the minimum expected cost with a probably approximately correct (PAC) guarantee on the annotation error, free of data distribution and pre-trained models. Experiments on common benchmarks demonstrate the effectiveness of our method, reducing the annotation cost by 78.51\\% while tightly controlling the annotation error.", "date": "2026-01-30", "categories": []}
{"id": "2503.09701", "title": "Reassessing Active Learning Adoption in Contemporary NLP: A Community Survey", "abstract": "Supervised learning relies on data annotation which usually is time-consuming and therefore expensive. A longstanding strategy to reduce annotation costs is active learning, an iterative process, in which a human annotates only data instances deemed informative by a model. Research in active learning has made considerable progress, especially with the rise of large language models (LLMs). However, we still know little about how these remarkable advances have translated into real-world applications, or contributed to removing key barriers to active learning adoption. To fill in this gap, we conduct an online survey in the NLP community to collect previously intangible insights on current implementation practices, common obstacles in application, and future prospects in active learning. We also reassess the perceived relevance of data annotation and active learning as fundamental assumptions. Our findings show that data annotation is expected to remain important and active learning to stay relevant while benefiting from LLMs. Consistent with a community survey from over 15 years ago, three key challenges yet persist -- setup complexity, uncertain cost reduction, and tooling -- for which we propose alleviation strategies. We publish an anonymized version of the dataset.", "date": "2025-03-12", "categories": []}
{"id": "2602.01581", "title": "Nearly Optimal Active Preference Learning and Its Application to LLM Alignment", "abstract": "Aligning large language models (LLMs) depends on high-quality datasets of human preference labels, which are costly to collect. Although active learning has been studied to improve sample efficiency relative to passive collection, many existing approaches adopt classical experimental design criteria such as G- or D-optimality. These objectives are not tailored to the structure of preference learning, leaving open the design of problem-specific algorithms. In this work, we identify a simple intuition specific to preference learning that calls into question the suitability of these existing design objectives. Motivated by this insight, we propose two active learning algorithms. The first provides the first instance-dependent label complexity guarantee for this setting, and the second is a simple, practical greedy method. We evaluate our algorithm on real-world preference datasets and observe improved sample efficiency compared to existing methods.", "date": "2026-02-02", "categories": []}
{"id": "2602.01898", "title": "Observation-dependent Bayesian active learning via input-warped Gaussian processes", "abstract": "Bayesian active learning relies on the precise quantification of predictive uncertainty to explore unknown function landscapes. While Gaussian process surrogates are the standard for such tasks, an underappreciated fact is that their posterior variance depends on the observed outputs only through the hyperparameters, rendering exploration largely insensitive to the actual measurements. We propose to inject observation-dependent feedback by warping the input space with a learned, monotone reparameterization. This mechanism allows the design policy to expand or compress regions of the input space in response to observed variability, thereby shaping the behavior of variance-based acquisition functions. We demonstrate that while such warps can be trained via marginal likelihood, a novel self-supervised objective yields substantially better performance. Our approach improves sample efficiency across a range of active learning benchmarks, particularly in regimes where non-stationarity challenges traditional methods.", "date": "2026-02-02", "categories": []}
{"id": "2602.02080", "title": "Learning Half-Spaces from Perturbed Contrastive Examples", "abstract": "We study learning under a two-step contrastive example oracle, as introduced by Mansouri et. al. (2025), where each queried (or sampled) labeled example is paired with an additional contrastive example of opposite label. While Mansouri et al. assume an idealized setting, where the contrastive example is at minimum distance of the originally queried/sampled point, we introduce and analyze a mechanism, parameterized by a non-decreasing noise function $f$, under which this ideal contrastive example is perturbed. The amount of perturbation is controlled by $f(d)$, where $d$ is the distance of the queried/sampled point to the decision boundary. Intuitively, this results in higher-quality contrastive examples for points closer to the decision boundary. We study this model in two settings: (i) when the maximum perturbation magnitude is fixed, and (ii) when it is stochastic.\n  For one-dimensional thresholds and for half-spaces under the uniform distribution on a bounded domain, we characterize active and passive contrastive sample complexity in dependence on the function $f$. We show that, under certain conditions on $f$, the presence of contrastive examples speeds up learning in terms of asymptotic query complexity and asymptotic expected query complexity.", "date": "2026-02-02", "categories": []}
{"id": "2602.02081", "title": "Active learning from positive and unlabeled examples", "abstract": "Learning from positive and unlabeled data (PU learning) is a weakly supervised variant of binary classification in which the learner receives labels only for (some) positively labeled instances, while all other examples remain unlabeled. Motivated by applications such as advertising and anomaly detection, we study an active PU learning setting where the learner can adaptively query instances from an unlabeled pool, but a queried label is revealed only when the instance is positive and an independent coin flip succeeds; otherwise the learner receives no information. In this paper, we provide the first theoretical analysis of the label complexity of active PU learning.", "date": "2026-02-02", "categories": []}
{"id": "2602.02415", "title": "Active Transfer Bagging: A New Approach for Accelerated Active Learning Acquisition of Data by Combined Transfer Learning and Bagging Based Models", "abstract": "Modern machine learning has achieved remarkable success on many problems, but this success often depends on the existence of large, labeled datasets. While active learning can dramatically reduce labeling cost when annotations are expensive, early performance is frequently dominated by the initial seed set, typically chosen at random. In many applications, however, related or approximate datasets are readily available and can be leveraged to construct a better seed set. We introduce a new method for selecting the seed data set for active learning, Active-Transfer Bagging (ATBagging). ATBagging estimates the informativeness of candidate data point from a Bayesian interpretation of bagged ensemble models by comparing in-bag and out-of-bag predictive distributions from the labeled dataset, yielding an information-gain proxy. To avoid redundant selections, we impose feature-space diversity by sampling a determinantal point process (DPP) whose kernel uses Random Fourier Features and a quality-diversity factorization that incorporates the informativeness scores. This same blended method is used for selection of new data points to collect during the active learning phase. We evaluate ATBagging on four real-world datasets covering both target-transfer and feature-shift scenarios (QM9, ERA5, Forbes 2000, and Beijing PM2.5). Across seed sizes nseed = 10-100, ATBagging improves or ties early active learning and increases area under the learning-curve relative to alternative seed subset selection methodologies in almost all cases, with strongest benefits in low-data regimes. Thus, ATBagging provides a low-cost, high reward means to initiating active learning-based data collection.", "date": "2026-02-02", "categories": []}
{"id": "2601.22732", "title": "Active Learning-Driven Lightweight YOLOv9: Enhancing Efficiency in Smart Agriculture", "abstract": "This study addresses the demand for real-time detection of tomatoes and tomato flowers by agricultural robots deployed on edge devices in greenhouse environments. Under practical imaging conditions, object detection systems often face challenges such as large scale variations caused by varying camera distances, severe occlusion from plant structures, and highly imbalanced class distributions. These factors make conventional object detection approaches that rely on fully annotated datasets difficult to simultaneously achieve high detection accuracy and deployment efficiency. To overcome these limitations, this research proposes an active learning driven lightweight object detection framework, integrating data analysis, model design, and training strategy. First, the size distribution of objects in raw agricultural images is analyzed to redefine an operational target range, thereby improving learning stability under real-world conditions. Second, an efficient feature extraction module is incorporated to reduce computational cost, while a lightweight attention mechanism is introduced to enhance feature representation under multi-scale and occluded scenarios. Finally, an active learning strategy is employed to iteratively select high-information samples for annotation and training under a limited labeling budget, effectively improving the recognition performance of minority and small-object categories. Experimental results demonstrate that, while maintaining a low parameter count and inference cost suitable for edge-device deployment, the proposed method effectively improves the detection performance of tomatoes and tomato flowers in raw images. Under limited annotation conditions, the framework achieves an overall detection accuracy of 67.8% mAP, validating its practicality and feasibility for intelligent agricultural applications.", "date": "2026-01-30", "categories": []}
{"id": "2601.21832", "title": "Goal-Driven Adaptive Sampling Strategies for Machine Learning Models Predicting Fields", "abstract": "Machine learning models are widely regarded as a way forward to tackle multi-query challenges that arise once expensive black-box simulations such as computational fluid dynamics are investigated. However, ensuring the desired level of accuracy for a certain task at minimal computational cost, e.g. as few black-box samples as possible, remains a challenges. Active learning strategies are used for scalar quantities to overcome this challenges and different so-called infill criteria exists and are commonly employed in several scenarios. Even though needed in various field an extension of active learning strategies towards field predictions is still lacking or limited to very specific scenarios and/or model types. In this paper we propose an active learning strategy for machine learning models that are capable if predicting field which is agnostic to the model architecture itself. For doing so, we combine a well-established Gaussian process model for a scalar reference value and simultaneously aim at reducing the epistemic model error and the difference between scalar and field predictions. Different specific forms of the above-mentioned approach are introduced and compared to each other as well as only scalar-valued based infill. Results are presented for the NASA common research model for an uncertainty propagation task showcasing high level of accuracy at significantly smaller cost compared to an approach without active learning.", "date": "2026-01-29", "categories": []}
{"id": "2601.21988", "title": "Generalized Information Gathering Under Dynamics Uncertainty", "abstract": "An agent operating in an unknown dynamical system must learn its dynamics from observations. Active information gathering accelerates this learning, but existing methods derive bespoke costs for specific modeling choices: dynamics models, belief update procedures, observation models, and planners. We present a unifying framework that decouples these choices from the information-gathering cost by explicitly exposing the causal dependencies between parameters, beliefs, and controls. Using this framework, we derive a general information-gathering cost based on Massey's directed information that assumes only Markov dynamics with additive noise and is otherwise agnostic to modeling choices. We prove that the mutual information cost used in existing literature is a special case of our cost. Then, we leverage our framework to establish an explicit connection between the mutual information cost and information gain in linearized Bayesian estimation, thereby providing theoretical justification for mutual information-based active learning approaches. Finally, we illustrate the practical utility of our framework through experiments spanning linear, nonlinear, and multi-agent systems.", "date": "2026-01-29", "categories": []}
{"id": "2601.20686", "title": "MuRAL-CPD: Active Learning for Multiresolution Change Point Detection", "abstract": "Change Point Detection (CPD) is a critical task in time series analysis, aiming to identify moments when the underlying data-generating process shifts. Traditional CPD methods often rely on unsupervised techniques, which lack adaptability to task-specific definitions of change and cannot benefit from user knowledge. To address these limitations, we propose MuRAL-CPD, a novel semi-supervised method that integrates active learning into a multiresolution CPD algorithm. MuRAL-CPD leverages a wavelet-based multiresolution decomposition to detect changes across multiple temporal scales and incorporates user feedback to iteratively optimize key hyperparameters. This interaction enables the model to align its notion of change with that of the user, improving both accuracy and interpretability. Our experimental results on several real-world datasets show the effectiveness of MuRAL-CPD against state-of-the-art methods, particularly in scenarios where minimal supervision is available.", "date": "2026-01-28", "categories": []}
{"id": "2601.17460", "title": "Entropy-Guided Agreement-Diversity: A Semi-Supervised Active Learning Framework for Fetal Head Segmentation in Ultrasound", "abstract": "Fetal ultrasound (US) data is often limited due to privacy and regulatory restrictions, posing challenges for training deep learning (DL) models. While semi-supervised learning (SSL) is commonly used for fetal US image analysis, existing SSL methods typically rely on random limited selection, which can lead to suboptimal model performance by overfitting to homogeneous labeled data. To address this, we propose a two-stage Active Learning (AL) sampler, Entropy-Guided Agreement-Diversity (EGAD), for fetal head segmentation. Our method first selects the most uncertain samples using predictive entropy, and then refines the final selection using the agreement-diversity score combining cosine similarity and mutual information. Additionally, our SSL framework employs a consistency learning strategy with feature downsampling to further enhance segmentation performance. In experiments, SSL-EGAD achieves an average Dice score of 94.57\\% and 96.32\\% on two public datasets for fetal head segmentation, using 5\\% and 10\\% labeled data for training, respectively. Our method outperforms current SSL models and showcases consistent robustness across diverse pregnancy stage data. The code is available on \\href{https://github.com/13204942/Semi-supervised-EGAD}{GitHub}.", "date": "2026-01-24", "categories": []}
{"id": "2601.15688", "title": "Performance-guided Reinforced Active Learning for Object Detection", "abstract": "Active learning (AL) strategies aim to train high-performance models with minimal labeling efforts, only selecting the most informative instances for annotation. Current approaches to evaluating data informativeness predominantly focus on the data's distribution or intrinsic information content and do not directly correlate with downstream task performance, such as mean average precision (mAP) in object detection. Thus, we propose Performance-guided (i.e. mAP-guided) Reinforced Active Learning for Object Detection (MGRAL), a novel approach that leverages the concept of expected model output changes as informativeness. To address the combinatorial explosion challenge of batch sample selection and the non-differentiable correlation between model performance and selected batches, MGRAL skillfully employs a reinforcement learning-based sampling agent that optimizes selection using policy gradient with mAP improvement as reward. Moreover, to reduce the computational overhead of mAP estimation with unlabeled samples, MGRAL utilizes an unsupervised way with fast look-up tables, ensuring feasible deployment. We evaluate MGRAL's active learning performance on detection tasks over PASCAL VOC and COCO benchmarks. Our approach demonstrates the highest AL curve with convincing visualizations, establishing a new paradigm in reinforcement learning-driven active object detection.", "date": "2026-01-22", "categories": []}
{"id": "2601.15773", "title": "Next Generation Active Learning: Mixture of LLMs in the Loop", "abstract": "With the rapid advancement and strong generalization capabilities of large language models (LLMs), they have been increasingly incorporated into the active learning pipelines as annotators to reduce annotation costs. However, considering the annotation quality, labels generated by LLMs often fall short of real-world applicability. To address this, we propose a novel active learning framework, Mixture of LLMs in the Loop Active Learning, replacing human annotators with labels generated through a Mixture-of-LLMs-based annotation model, aimed at enhancing LLM-based annotation robustness by aggregating the strengths of multiple LLMs. To further mitigate the impact of the noisy labels, we introduce annotation discrepancy and negative learning to identify the unreliable annotations and enhance learning effectiveness. Extensive experiments demonstrate that our framework achieves performance comparable to human annotation and consistently outperforms single-LLM baselines and other LLM-ensemble-based approaches. Moreover, our framework is built on lightweight LLMs, enabling it to operate fully on local machines in real-world applications.", "date": "2026-01-22", "categories": []}
{"id": "2601.06916", "title": "Active Learning Strategies for Efficient Machine-Learned Interatomic Potentials Across Diverse Material Systems", "abstract": "Efficient materials discovery requires reducing costly first-principles calculations for training machine-learned interatomic potentials (MLIPs). We develop an active learning (AL) framework that iteratively selects informative structures from the Materials Project and Open Quantum Materials Database (OQMD) using compositional and property-based descriptors with a neural network ensemble model. Query-by-Committee enables real-time uncertainty quantification. We compare four strategies: random sampling (baseline), uncertainty-based sampling, diversity-based sampling (k-means clustering with farthest-point refinement), and a hybrid approach. Experiments across four material systems (C, Si, Fe, and TiO2) with 5 random seeds demonstrate that diversity sampling achieves competitive or superior performance, with 10.9% improvement on TiO2. Our approach achieves equivalent accuracy with 5-13% fewer labeled samples than random baselines. The complete pipeline executes on Google Colab in under 4 hours per system using less than 8 GB RAM, democratizing MLIP development for resource-limited researchers. Open-source code and configurations are available on GitHub. This multi-system evaluation provides practical guidelines for data-efficient MLIP training and highlights integration with symmetry-aware architectures as a promising future direction.", "date": "2026-01-11", "categories": []}
{"id": "2601.10993", "title": "Memorize Early, Then Query: Inlier-Memorization-Guided Active Outlier Detection", "abstract": "Outlier detection (OD) aims to identify abnormal instances, known as outliers or anomalies, by learning typical patterns of normal data, or inliers. Performing OD under an unsupervised regime-without any information about anomalous instances in the training data-is challenging. A recently observed phenomenon, known as the inlier-memorization (IM) effect, where deep generative models (DGMs) tend to memorize inlier patterns during early training, provides a promising signal for distinguishing outliers. However, existing unsupervised approaches that rely solely on the IM effect still struggle when inliers and outliers are not well-separated or when outliers form dense clusters. To address these limitations, we incorporate active learning to selectively acquire informative labels, and propose IMBoost, a novel framework that explicitly reinforces the IM effect to improve outlier detection. Our method consists of two stages: 1) a warm-up phase that induces and promotes the IM effect, and 2) a polarization phase in which actively queried samples are used to maximize the discrepancy between inlier and outlier scores. In particular, we propose a novel query strategy and tailored loss function in the polarization phase to effectively identify informative samples and fully leverage the limited labeling budget. We provide a theoretical analysis showing that the IMBoost consistently decreases inlier risk while increasing outlier risk throughout training, thereby amplifying their separation. Extensive experiments on diverse benchmark datasets demonstrate that IMBoost not only significantly outperforms state-of-the-art active OD methods but also requires substantially less computational cost.", "date": "2026-01-16", "categories": []}
{"id": "2601.13677", "title": "Finally Outshining the Random Baseline: A Simple and Effective Solution for Active Learning in 3D Biomedical Imaging", "abstract": "Active learning (AL) has the potential to drastically reduce annotation costs in 3D biomedical image segmentation, where expert labeling of volumetric data is both time-consuming and expensive. Yet, existing AL methods are unable to consistently outperform improved random sampling baselines adapted to 3D data, leaving the field without a reliable solution. We introduce Class-stratified Scheduled Power Predictive Entropy (ClaSP PE), a simple and effective query strategy that addresses two key limitations of standard uncertainty-based AL methods: class imbalance and redundancy in early selections. ClaSP PE combines class-stratified querying to ensure coverage of underrepresented structures and log-scale power noising with a decaying schedule to enforce query diversity in early-stage AL and encourage exploitation later. In our evaluation on 24 experimental settings using four 3D biomedical datasets within the comprehensive nnActive benchmark, ClaSP PE is the only method that generally outperforms improved random baselines in terms of both segmentation quality with statistically significant gains, whilst remaining annotation efficient. Furthermore, we explicitly simulate the real-world application by testing our method on four previously unseen datasets without manual adaptation, where all experiment parameters are set according to predefined guidelines. The results confirm that ClaSP PE robustly generalizes to novel tasks without requiring dataset-specific tuning. Within the nnActive framework, we present compelling evidence that an AL method can consistently outperform random baselines adapted to 3D segmentation, in terms of both performance and annotation efficiency in a realistic, close-to-production scenario. Our open-source implementation and clear deployment guidelines make it readily applicable in practice. Code is at https://github.com/MIC-DKFZ/nnActive.", "date": "2026-01-20", "categories": []}
{"id": "2405.10808", "title": "ActiveLLM: Large Language Model-based Active Learning for Textual Few-Shot Scenarios", "abstract": "Active learning is designed to minimize annotation efforts by prioritizing instances that most enhance learning. However, many active learning strategies struggle with a `cold-start' problem, needing substantial initial data to be effective. This limitation reduces their utility in the increasingly relevant few-shot scenarios, where the instance selection has a substantial impact. To address this, we introduce ActiveLLM, a novel active learning approach that leverages Large Language Models such as GPT-4, o1, Llama 3, or Mistral Large for selecting instances. We demonstrate that ActiveLLM significantly enhances the classification performance of BERT classifiers in few-shot scenarios, outperforming traditional active learning methods as well as improving the few-shot learning methods ADAPET, PERFECT, and SetFit. Additionally, ActiveLLM can be extended to non-few-shot scenarios, allowing for iterative selections. In this way, ActiveLLM can even help other active learning strategies to overcome their cold-start problem. Our results suggest that ActiveLLM offers a promising solution for improving model performance across various learning setups.", "date": "2024-05-17", "categories": []}
{"id": "2601.08024", "title": "A Highly Efficient Diversity-based Input Selection for DNN Improvement Using VLMs", "abstract": "Maintaining or improving the performance of Deep Neural Networks (DNNs) through fine-tuning requires labeling newly collected inputs, a process that is often costly and time-consuming. To alleviate this problem, input selection approaches have been developed in recent years to identify small, yet highly informative subsets for labeling. Diversity-based selection is one of the most effective approaches for this purpose. However, they are often computationally intensive and lack scalability for large input sets, limiting their practical applicability. To address this challenge, we introduce Concept-Based Diversity (CBD), a highly efficient metric for image inputs that leverages Vision-Language Models (VLM). Our results show that CBD exhibits a strong correlation with Geometric Diversity (GD), an established diversity metric, while requiring only a fraction of its computation time. Building on this finding, we propose a hybrid input selection approach that combines CBD with Margin, a simple uncertainty metric. We conduct a comprehensive evaluation across a diverse set of DNN models, input sets, selection budgets, and five most effective state-of-the-art selection baselines. The results demonstrate that the CBD-based selection consistently outperforms all baselines at guiding input selection to improve the DNN model. Furthermore, the CBD-based selection approach remains highly efficient, requiring selection times close to those of simple uncertainty-based methods such as Margin, even on larger input sets like ImageNet. These results confirm not only the effectiveness and computational advantage of the CBD-based approach, particularly compared to hybrid baselines, but also its scalability in repetitive and extensive input selection scenarios.", "date": "2026-01-12", "categories": []}
{"id": "2601.06209", "title": "When Imbalance Comes Twice: Active Learning under Simulated Class Imbalance and Label Shift in Binary Semantic Segmentation", "abstract": "The aim of Active Learning is to select the most informative samples from an unlabelled set of data. This is useful in cases where the amount of data is large and labelling is expensive, such as in machine vision or medical imaging. Two particularities of machine vision are first, that most of the images produced are free of defects, and second, that the amount of images produced is so big that we cannot store all acquired images. This results, on the one hand, in a strong class imbalance in defect distribution and, on the other hand, in a potential label shift caused by limited storage. To understand how these two forms of imbalance affect active learning algorithms, we propose a simulation study based on two open-source datasets. We artificially create datasets for which we control the levels of class imbalance and label shift. Three standard active learning selection strategies are compared: random sampling, entropy-based selection, and core-set selection. We demonstrate that active learning strategies, and in particular the entropy-based and core-set selections, remain interesting and efficient even for highly imbalanced datasets. We also illustrate and measure the loss of efficiency that occurs in the situation a strong label shift.", "date": "2026-01-08", "categories": []}
{"id": "2506.05999", "title": "Machine learning for in-situ composition mapping in a self-driving magnetron sputtering system", "abstract": "Self-driving labs (SDLs), employing automation and machine learning (ML) to accelerate experimental procedures, have enormous potential in the discovery of new materials. However, in thin film science, SDLs are mainly restricted to solution-based synthetic methods which are easier to automate but cannot access the broad chemical space of inorganic materials. This work presents an SDL based on magnetron co-sputtering. We are using combinatorial frameworks, obtaining accurate composition maps on multi-element, compositionally graded thin films. This normally requires time-consuming ex-situ analysis prone to systematic errors. We present a rapid and calibration-free in-situ, ML driven approach to produce composition maps for arbitrary source combinations and sputtering conditions. We develop a method to predict the composition distribution in a multi-element combinatorial thin film, using in-situ measurements from quartz-crystal microbalance sensors placed in a sputter chamber. For a given source, the sensor readings are learned as a function of the sputtering pressure and magnetron power, through active learning using Gaussian processes (GPs). The final GPs are combined with a geometric model of the deposition flux distribution in the chamber, which allows interpolation of the deposition rates from each source, at any position across the sample. We investigate several acquisition functions for the ML procedure. A fully Bayesian GP - BALM (Bayesian active learning MacKay) - achieved the best performance, learning the deposition rates for a single source in 10 experiments. Prediction accuracy for co-sputtering composition distributions was verified experimentally. Our framework dramatically increases throughput by avoiding the need for extensive characterisation or calibration, thus demonstrating the potential of ML-guided SDLs to accelerate materials exploration.", "date": "2025-06-06", "categories": []}
{"id": "2601.05946", "title": "A Critical Examination of Active Learning Workflows in Materials Science", "abstract": "Active learning (AL) plays a critical role in materials science, enabling applications such as the construction of machine-learning interatomic potentials for atomistic simulations and the operation of self-driving laboratories. Despite its widespread use, the reliability and effectiveness of AL workflows depend on implicit design assumptions that are rarely examined systematically. Here, we critically assess AL workflows deployed in materials science and investigate how key design choices, such as surrogate models, sampling strategies, uncertainty quantification and evaluation metrics, relate to their performance. By identifying common pitfalls and discussing practical mitigation strategies, we provide guidance to practitioners for the efficient design, assessment, and interpretation of AL workflows in materials science.", "date": "2026-01-09", "categories": []}
{"id": "2601.01695", "title": "Learnability-Driven Submodular Optimization for Active Roadside 3D Detection", "abstract": "Roadside perception datasets are typically constructed via cooperative labeling between synchronized vehicle and roadside frame pairs. However, real deployment often requires annotation of roadside-only data due to hardware and privacy constraints. Even human experts struggle to produce accurate labels without vehicle-side data (image, LIDAR), which not only increases annotation difficulty and cost, but also reveals a fundamental learnability problem: many roadside-only scenes contain distant, blurred, or occluded objects whose 3D properties are ambiguous from a single view and can only be reliably annotated by cross-checking paired vehicle--roadside frames. We refer to such cases as inherently ambiguous samples. To reduce wasted annotation effort on inherently ambiguous samples while still obtaining high-performing models, we turn to active learning. This work focuses on active learning for roadside monocular 3D object detection and proposes a learnability-driven framework that selects scenes which are both informative and reliably labelable, suppressing inherently ambiguous samples while ensuring coverage. Experiments demonstrate that our method, LH3D, achieves 86.06%, 67.32%, and 78.67% of full-performance for vehicles, pedestrians, and cyclists respectively, using only 25% of the annotation budget on DAIR-V2X-I, significantly outperforming uncertainty-based baselines. This confirms that learnability, not uncertainty, matters for roadside 3D perception.", "date": "2026-01-04", "categories": []}
{"id": "2210.08367", "title": "Active Learning with Neural Networks: Insights from Nonparametric Statistics", "abstract": "Deep neural networks have great representation power, but typically require large numbers of training examples. This motivates deep active learning methods that can significantly reduce the amount of labeled training data. Empirical successes of deep active learning have been recently reported in the literature, however, rigorous label complexity guarantees of deep active learning have remained elusive. This constitutes a significant gap between theory and practice. This paper tackles this gap by providing the first near-optimal label complexity guarantees for deep active learning. The key insight is to study deep active learning from the nonparametric classification perspective. Under standard low noise conditions, we show that active learning with neural networks can provably achieve the minimax label complexity, up to disagreement coefficient and other logarithmic terms. When equipped with an abstention option, we further develop an efficient deep active learning algorithm that achieves $\\mathsf{polylog}(\\frac{1}{\\epsilon})$ label complexity, without any low noise assumptions. We also provide extensions of our results beyond the commonly studied Sobolev/H\\\"older spaces and develop label complexity guarantees for learning in Radon $\\mathsf{BV}^2$ spaces, which have recently been proposed as natural function spaces associated with neural networks.", "date": "2022-10-15", "categories": []}
{"id": "2512.20021", "title": "Gaussian Process Assisted Meta-learning for Image Classification and Object Detection Models", "abstract": "Collecting operationally realistic data to inform machine learning models can be costly. Before collecting new data, it is helpful to understand where a model is deficient. For example, object detectors trained on images of rare objects may not be good at identification in poorly represented conditions. We offer a way of informing subsequent data acquisition to maximize model performance by leveraging the toolkit of computer experiments and metadata describing the circumstances under which the training data was collected (e.g., season, time of day, location). We do this by evaluating the learner as the training data is varied according to its metadata. A Gaussian process (GP) surrogate fit to that response surface can inform new data acquisitions. This meta-learning approach offers improvements to learner performance as compared to data with randomly selected metadata, which we illustrate on both classic learning examples, and on a motivating application involving the collection of aerial images in search of airplanes.", "date": "2025-12-23", "categories": []}
{"id": "2510.21052", "title": "Amortized Active Generation of Pareto Sets", "abstract": "We introduce active generation of Pareto sets (A-GPS), a new framework for online discrete black-box multi-objective optimization (MOO). A-GPS learns a generative model of the Pareto set that supports a-posteriori conditioning on user preferences. The method employs a class probability estimator (CPE) to predict non-dominance relations and to condition the generative model toward high-performing regions of the search space. We also show that this non-dominance CPE implicitly estimates the probability of hypervolume improvement (PHVI). To incorporate subjective trade-offs, A-GPS introduces preference direction vectors that encode user-specified preferences in objective space. At each iteration, the model is updated using both Pareto membership and alignment with these preference directions, producing an amortized generative model capable of sampling across the Pareto front without retraining. The result is a simple yet powerful approach that achieves high-quality Pareto set approximations, avoids explicit hypervolume computation, and flexibly captures user preferences. Empirical results on synthetic benchmarks and protein design tasks demonstrate strong sample efficiency and effective preference incorporation.", "date": "2025-10-23", "categories": []}
{"id": "2512.14170", "title": "On Improving Deep Active Learning with Formal Verification", "abstract": "Deep Active Learning (DAL) aims to reduce labeling costs in neural-network training by prioritizing the most informative unlabeled samples for annotation. Beyond selecting which samples to label, several DAL approaches further enhance data efficiency by augmenting the training set with synthetic inputs that do not require additional manual labeling. In this work, we investigate how augmenting the training data with adversarial inputs that violate robustness constraints can improve DAL performance. We show that adversarial examples generated via formal verification contribute substantially more than those produced by standard, gradient-based attacks. We apply this extension to multiple modern DAL techniques, as well as to a new technique that we propose, and show that it yields significant improvements in model generalization across standard benchmarks.", "date": "2025-12-16", "categories": []}
{"id": "2510.23681", "title": "Informed Initialization for Bayesian Optimization and Active Learning", "abstract": "Bayesian Optimization is a widely used method for optimizing expensive black-box functions, relying on probabilistic surrogate models such as Gaussian Processes. The quality of the surrogate model is crucial for good optimization performance, especially in the few-shot setting where only a small number of batches of points can be evaluated. In this setting, the initialization plays a critical role in shaping the surrogate's predictive quality and guiding subsequent optimization. Despite this, practitioners typically rely on (quasi-)random designs to cover the input space. However, such approaches neglect two key factors: (a) space-filling designs may not be desirable to reduce predictive uncertainty, and (b) efficient hyperparameter learning during initialization is essential for high-quality prediction, which may conflict with space-filling designs. To address these limitations, we propose Hyperparameter-Informed Predictive Exploration (HIPE), a novel acquisition strategy that balances predictive uncertainty reduction with hyperparameter learning using information-theoretic principles. We derive a closed-form expression for HIPE in the Gaussian Process setting and demonstrate its effectiveness through extensive experiments in active learning and few-shot BO. Our results show that HIPE outperforms standard initialization strategies in terms of predictive accuracy, hyperparameter identification, and subsequent optimization performance, particularly in large-batch, few-shot settings relevant to many real-world Bayesian Optimization applications.", "date": "2025-10-27", "categories": []}
{"id": "2512.12870", "title": "Optimal Labeler Assignment and Sampling for Active Learning in the Presence of Imperfect Labels", "abstract": "Active Learning (AL) has garnered significant interest across various application domains where labeling training data is costly. AL provides a framework that helps practitioners query informative samples for annotation by oracles (labelers). However, these labels often contain noise due to varying levels of labeler accuracy. Additionally, uncertain samples are more prone to receiving incorrect labels because of their complexity. Learning from imperfectly labeled data leads to an inaccurate classifier. We propose a novel AL framework to construct a robust classification model by minimizing noise levels. Our approach includes an assignment model that optimally assigns query points to labelers, aiming to minimize the maximum possible noise within each cycle. Additionally, we introduce a new sampling method to identify the best query points, reducing the impact of label noise on classifier performance. Our experiments demonstrate that our approach significantly improves classification performance compared to several benchmark methods.", "date": "2025-12-14", "categories": []}
{"id": "2512.13530", "title": "Actively Learning Joint Contours of Multiple Computer Experiments", "abstract": "Contour location$\\unicode{x2014}$the process of sequentially training a surrogate model to identify the design inputs that result in a pre-specified response value from a single computer experiment$\\unicode{x2014}$is a well-studied active learning problem. Here, we tackle a related but distinct problem: identifying the input configuration that returns pre-specified values of multiple independent computer experiments simultaneously. Motivated by computer experiments of the rotational torques acting upon a vehicle in flight, we aim to identify stable flight conditions which result in zero torque forces. We propose a \"joint contour location\" (jCL) scheme that strikes a strategic balance between exploring the multiple response surfaces while exploiting learning of the intersecting contours. We employ both shallow and deep Gaussian process surrogates, but our jCL procedure is applicable to any surrogate that can provide posterior predictive distributions. Our jCL designs significantly outperform existing (single response) CL strategies, enabling us to efficiently locate the joint contour of our motivating computer experiments.", "date": "2025-12-15", "categories": []}
{"id": "2402.03587", "title": "Information-Theoretic Active Correlation Clustering", "abstract": "Correlation clustering is a flexible framework for partitioning data based solely on pairwise similarity or dissimilarity information, without requiring the number of clusters as input. However, in many practical scenarios, these pairwise similarities are not available a priori and must be obtained through costly measurements or human feedback. This motivates the use of active learning to query only the most informative pairwise comparisons, enabling effective clustering under budget constraints. In this work, we develop a principled active learning approach for correlation clustering by introducing several information-theoretic acquisition functions that prioritize queries based on entropy and expected information gain. These strategies aim to reduce uncertainty about the clustering structure as efficiently as possible. We evaluate our methods across a range of synthetic and real-world settings and show that they significantly outperform existing baselines in terms of clustering accuracy and query efficiency. Our results highlight the benefits of combining active learning with correlation clustering in settings where similarity information is costly or limited.", "date": "2024-02-05", "categories": []}
{"id": "2512.07606", "title": "Decomposition Sampling for Efficient Region Annotations in Active Learning", "abstract": "Active learning improves annotation efficiency by selecting the most informative samples for annotation and model training. While most prior work has focused on selecting informative images for classification tasks, we investigate the more challenging setting of dense prediction, where annotations are more costly and time-intensive, especially in medical imaging. Region-level annotation has been shown to be more efficient than image-level annotation for these tasks. However, existing methods for representative annotation region selection suffer from high computational and memory costs, irrelevant region choices, and heavy reliance on uncertainty sampling. We propose decomposition sampling (DECOMP), a new active learning sampling strategy that addresses these limitations. It enhances annotation diversity by decomposing images into class-specific components using pseudo-labels and sampling regions from each class. Class-wise predictive confidence further guides the sampling process, ensuring that difficult classes receive additional annotations. Across ROI classification, 2-D segmentation, and 3-D segmentation, DECOMP consistently surpasses baseline methods by better sampling minority-class regions and boosting performance on these challenging classes. Code is in https://github.com/JingnaQiu/DECOMP.git.", "date": "2025-12-08", "categories": []}
{"id": "2511.19730", "title": "Training-Free Active Learning Framework in Materials Science with Large Language Models", "abstract": "Active learning (AL) accelerates scientific discovery by prioritizing the most informative experiments, but traditional machine learning (ML) models used in AL suffer from cold-start limitations and domain-specific feature engineering, restricting their generalizability. Large language models (LLMs) offer a new paradigm by leveraging their pretrained knowledge and universal token-based representations to propose experiments directly from text-based descriptions. Here, we introduce an LLM-based active learning framework (LLM-AL) that operates in an iterative few-shot setting and benchmark it against conventional ML models across four diverse materials science datasets. We explored two prompting strategies: one using concise numerical inputs suited for datasets with more compositional and structured features, and another using expanded descriptive text suited for datasets with more experimental and procedural features to provide additional context. Across all datasets, LLM-AL could reduce the number of experiments needed to reach top-performing candidates by over 70% and consistently outperformed traditional ML models. We found that LLM-AL performs broader and more exploratory searches while still reaching the optima with fewer iterations. We further examined the stability boundaries of LLM-AL given the inherent non-determinism of LLMs and found its performance to be broadly consistent across runs, within the variability range typically observed for traditional ML approaches. These results demonstrate that LLM-AL can serve as a generalizable alternative to conventional AL pipelines for more efficient and interpretable experiment selection and potential LLM-driven autonomous discovery.", "date": "2025-11-24", "categories": []}
{"id": "2506.11599", "title": "A$^2$LC: Active and Automated Label Correction for Semantic Segmentation", "abstract": "Active Label Correction (ALC) has emerged as a promising solution to the high cost and error-prone nature of manual pixel-wise annotation in semantic segmentation, by actively identifying and correcting mislabeled data. Although recent work has improved correction efficiency by generating pseudo-labels using foundation models, substantial inefficiencies still remain. In this paper, we introduce A$^2$LC, an Active and Automated Label Correction framework for semantic segmentation, where manual and automatic correction stages operate in a cascaded manner. Specifically, the automatic correction stage leverages human feedback to extend label corrections beyond the queried samples, thereby maximizing cost efficiency. In addition, we introduce an adaptively balanced acquisition function that emphasizes underrepresented tail classes, working in strong synergy with the automatic correction stage. Extensive experiments on Cityscapes and PASCAL VOC 2012 demonstrate that A$^2$LC significantly outperforms previous state-of-the-art methods. Notably, A$^2$LC exhibits high efficiency by outperforming previous methods with only 20% of their budget, and shows strong effectiveness by achieving a 27.23% performance gain under the same budget on Cityscapes.", "date": "2025-06-13", "categories": []}
{"id": "2512.01783", "title": "The Active and Noise-Tolerant Strategic Perceptron", "abstract": "We initiate the study of active learning algorithms for classifying strategic agents. Active learning is a well-established framework in machine learning in which the learner selectively queries labels, often achieving substantially higher accuracy and efficiency than classical supervised methods-especially in settings where labeling is costly or time-consuming, such as hiring, admissions, and loan decisions. Strategic classification, however, addresses scenarios where agents modify their features to obtain more favorable outcomes, resulting in observed data that is not truthful. Such manipulation introduces challenges beyond those in learning from clean data. Our goal is to design active and noise-tolerant algorithms that remain effective in strategic environments-algorithms that classify strategic agents accurately while issuing as few label requests as possible. The central difficulty is to simultaneously account for strategic manipulation and preserve the efficiency gains of active learning.\n  Our main result is an algorithm for actively learning linear separators in the strategic setting that preserves the exponential improvement in label complexity over passive learning previously obtained only in the non-strategic case. Specifically, for data drawn uniformly from the unit sphere, we show that a modified version of the Active Perceptron algorithm [DKM05,YZ17] achieves excess error $\\epsilon$ using only $\\tilde{O}(d \\ln \\frac{1}{\\epsilon})$ label queries and incurs at most $\\tilde{O}(d \\ln \\frac{1}{\\epsilon})$ additional mistakes relative to the optimal classifier, even in the nonrealizable case, when a $\\tilde{\\Omega}(\\epsilon)$ fraction of inputs have inconsistent labels with the optimal classifier. The algorithm is computationally efficient and, under these distributional assumptions, requires substantially fewer label queries than prior work on strategic Perceptron [ABBN21].", "date": "2025-12-01", "categories": []}
{"id": "2411.18254", "title": "Active partitioning: inverting the paradigm of active learning", "abstract": "Datasets often incorporate various functional patterns related to different aspects or regimes, which are typically not equally present throughout the dataset. We propose a novel, general-purpose partitioning algorithm that utilizes competition between models to detect and separate these functional patterns. This competition is induced by multiple models iteratively submitting their predictions for the dataset, with the best prediction for each data point being rewarded with training on that data point. This reward mechanism amplifies each model's strengths and encourages specialization in different patterns. The specializations can then be translated into a partitioning scheme. The amplification of each model's strengths inverts the active learning paradigm: while active learning typically focuses the training of models on their weaknesses to minimize the number of required training data points, our concept reinforces the strengths of each model, thus specializing them. We validate our concept -- called active partitioning -- with various datasets with clearly distinct functional patterns, such as mechanical stress and strain data in a porous structure. The active partitioning algorithm produces valuable insights into the datasets' structure, which can serve various further applications. As a demonstration of one exemplary usage, we set up modular models consisting of multiple expert models, each learning a single partition, and compare their performance on more than twenty popular regression problems with single models learning all partitions simultaneously. Our results show significant improvements, with up to 54% loss reduction, confirming our partitioning algorithm's utility.", "date": "2024-11-27", "categories": []}
{"id": "2511.18058", "title": "Hierarchical Semi-Supervised Active Learning for Remote Sensing", "abstract": "The performance of deep learning models in remote sensing (RS) strongly depends on the availability of high-quality labeled data. However, collecting large-scale annotations is costly and time-consuming, while vast amounts of unlabeled imagery remain underutilized. To address this challenge, we propose a Hierarchical Semi-Supervised Active Learning (HSSAL) framework that integrates semi-supervised learning (SSL) and a novel hierarchical active learning (HAL) in a closed iterative loop. In each iteration, SSL refines the model using both labeled data through supervised learning and unlabeled data via weak-to-strong self-training, improving feature representation and uncertainty estimation. Guided by the refined representations and uncertainty cues of unlabeled samples, HAL then conducts sample querying through a progressive clustering strategy, selecting the most informative instances that jointly satisfy the criteria of scalability, diversity, and uncertainty. This hierarchical process ensures both efficiency and representativeness in sample selection. Extensive experiments on three benchmark RS scene classification datasets, including UCM, AID, and NWPU-RESISC45, demonstrate that HSSAL consistently outperforms SSL- or AL-only baselines. Remarkably, with only 8%, 4%, and 2% labeled training data on UCM, AID, and NWPU-RESISC45, respectively, HSSAL achieves over 95% of fully-supervised accuracy, highlighting its superior label efficiency through informativeness exploitation of unlabeled data. Our code will be publicly available.", "date": "2025-11-22", "categories": []}
{"id": "2512.00453", "title": "Sample-Efficient Expert Query Control in Active Imitation Learning via Conformal Prediction", "abstract": "Active imitation learning (AIL) combats covariate shift by querying an expert during training. However, expert action labeling often dominates the cost, especially in GPU-intensive simulators, human-in-the-loop settings, and robot fleets that revisit near-duplicate states. We present Conformalized Rejection Sampling for Active Imitation Learning (CRSAIL), a querying rule that requests an expert action only when the visited state is under-represented in the expert-labeled dataset. CRSAIL scores state novelty by the distance to the $K$-th nearest expert state and sets a single global threshold via conformal prediction. This threshold is the empirical $(1-\\alpha)$ quantile of on-policy calibration scores, providing a distribution-free calibration rule that links $\\alpha$ to the expected query rate and makes $\\alpha$ a task-agnostic tuning knob. This state-space querying strategy is robust to outliers and, unlike safety-gate-based AIL, can be run without real-time expert takeovers: we roll out full trajectories (episodes) with the learner and only afterward query the expert on a subset of visited states. Evaluated on MuJoCo robotics tasks, CRSAIL matches or exceeds expert-level reward while reducing total expert queries by up to 96% vs. DAgger and up to 65% vs. prior AIL methods, with empirical robustness to $\\alpha$ and $K$, easing deployment on novel systems with unknown dynamics.", "date": "2025-11-29", "categories": []}
{"id": "2211.10360", "title": "Data efficient surrogate modeling for engineering design: Ensemble-free batch mode deep active learning for regression", "abstract": "High fidelity design evaluation processes such as Computational Fluid Dynamics and Finite Element Analysis are often replaced with data driven surrogates to reduce computational cost in engineering design optimization. However, building accurate surrogate models still requires a large number of expensive simulations. To address this challenge, we introduce epsilon HQS, a scalable active learning strategy that leverages a student teacher framework to train deep neural networks efficiently. Unlike Bayesian AL methods, which are computationally demanding with DNNs, epsilon HQS selectively queries informative samples to reduce labeling cost. Applied to CFD, FEA, and propeller design tasks, our method achieves higher accuracy under fixed labeling cost budgets.", "date": "2022-11-16", "categories": []}
{"id": "2511.22567", "title": "Where to Measure: Epistemic Uncertainty-Based Sensor Placement with ConvCNPs", "abstract": "Accurate sensor placement is critical for modeling spatio-temporal systems such as environmental and climate processes. Neural Processes (NPs), particularly Convolutional Conditional Neural Processes (ConvCNPs), provide scalable probabilistic models with uncertainty estimates, making them well-suited for data-driven sensor placement. However, existing approaches rely on total predictive uncertainty, which conflates epistemic and aleatoric components, that may lead to suboptimal sensor selection in ambiguous regions. To address this, we propose expected reduction in epistemic uncertainty as a new acquisition function for sensor placement. To enable this, we extend ConvCNPs with a Mixture Density Networks (MDNs) output head for epistemic uncertainty estimation. Preliminary results suggest that epistemic uncertainty driven sensor placement more effectively reduces model error than approaches based on overall uncertainty.", "date": "2025-11-27", "categories": []}
{"id": "2504.16136", "title": "Active Learning Methods for Efficient Data Utilization and Model Performance Enhancement", "abstract": "In the era of data-driven intelligence, the paradox of data abundance and annotation scarcity has emerged as a critical bottleneck in the advancement of machine learning. This paper gives a detailed overview of Active Learning (AL), which is a strategy in machine learning that helps models achieve better performance using fewer labeled examples. It introduces the basic concepts of AL and discusses how it is used in various fields such as computer vision, natural language processing, transfer learning, and real-world applications. The paper focuses on important research topics such as uncertainty estimation, handling of class imbalance, domain adaptation, fairness, and the creation of strong evaluation metrics and benchmarks. It also shows that learning methods inspired by humans and guided by questions can improve data efficiency and help models learn more effectively. In addition, this paper talks about current challenges in the field, including the need to rebuild trust, ensure reproducibility, and deal with inconsistent methodologies. It points out that AL often gives better results than passive learning, especially when good evaluation measures are used. This work aims to be useful for both researchers and practitioners by providing key insights and proposing directions for future progress in active learning.", "date": "2025-04-21", "categories": []}
{"id": "2508.15577", "title": "LFaB: Low fidelity as Bias for Active Learning in the chemical configuration space", "abstract": "Active learning promises to provide an optimal training sample selection procedure in the construction of machine learning models. It often relies on minimizing the model's variance, which is assumed to decrease the prediction error. Still, it is frequently even less efficient than pure random sampling. Motivated by the bias-variance decomposition, we propose to minimize the model's bias instead of its variance. By doing so, we are able to almost exactly match the best-case error over all possible greedy sample selection procedures for a relevant application. Our bias approximation is based on using cheap to calculate low fidelity data as known from $\\Delta$-ML or multifidelity machine learning. We exemplify our approach for a wider class of applications in quantum chemistry including predicting excitation energies and ab initio potential energy surfaces. Here, the proposed method reduces training data consumption by up to an order of magnitude compared to standard active learning.", "date": "2025-08-21", "categories": []}
{"id": "2509.10390", "title": "Vendi Information Gain for Active Learning and its Application to Ecology", "abstract": "While monitoring biodiversity through camera traps has become an important endeavor for ecological research, identifying species in the captured image data remains a major bottleneck due to limited labeling resources. Active learning -- a machine learning paradigm that selects the most informative data to label and train a predictive model -- offers a promising solution, but typically focuses on uncertainty in the individual predictions without considering uncertainty across the entire dataset. We introduce a new active learning policy, Vendi information gain (VIG), that selects images based on their impact on dataset-wide prediction uncertainty, capturing both informativeness and diversity. We applied VIG to the Snapshot Serengeti dataset and compared it against common active learning methods. VIG needs only 3% of the available data to reach 75% accuracy, a level that baselines require more than 10% of the data to achieve. With 10% of the data, VIG attains 88% predictive accuracy, 12% higher than the best of the baselines. This improvement in performance is consistent across metrics and batch sizes, and we show that VIG also collects more diverse data in the feature space. VIG has broad applicability beyond ecology, and our results highlight its value for biodiversity monitoring in data-limited environments.", "date": "2025-09-12", "categories": []}
{"id": "2511.07473", "title": "RELEAP: Reinforcement-Enhanced Label-Efficient Active Phenotyping for Electronic Health Records", "abstract": "Objective: Electronic health record (EHR) phenotyping often relies on noisy proxy labels, which undermine the reliability of downstream risk prediction. Active learning can reduce annotation costs, but most rely on fixed heuristics and do not ensure that phenotype refinement improves prediction performance. Our goal was to develop a framework that directly uses downstream prediction performance as feedback to guide phenotype correction and sample selection under constrained labeling budgets.\n  Materials and Methods: We propose Reinforcement-Enhanced Label-Efficient Active Phenotyping (RELEAP), a reinforcement learning-based active learning framework. RELEAP adaptively integrates multiple querying strategies and, unlike prior methods, updates its policy based on feedback from downstream models. We evaluated RELEAP on a de-identified Duke University Health System (DUHS) cohort (2014-2024) for incident lung cancer risk prediction, using logistic regression and penalized Cox survival models. Performance was benchmarked against noisy-label baselines and single-strategy active learning.\n  Results: RELEAP consistently outperformed all baselines. Logistic AUC increased from 0.774 to 0.805 and survival C-index from 0.718 to 0.752. Using downstream performance as feedback, RELEAP produced smoother and more stable gains than heuristic methods under the same labeling budget.\n  Discussion: By linking phenotype refinement to prediction outcomes, RELEAP learns which samples most improve downstream discrimination and calibration, offering a more principled alternative to fixed active learning rules.\n  Conclusion: RELEAP optimizes phenotype correction through downstream feedback, offering a scalable, label-efficient paradigm that reduces manual chart review and enhances the reliability of EHR-based risk prediction.", "date": "2025-11-08", "categories": []}
{"id": "2511.17760", "title": "When Active Learning Fails, Uncalibrated Out of Distribution Uncertainty Quantification Might Be the Problem", "abstract": "Efficiently and meaningfully estimating prediction uncertainty is important for exploration in active learning campaigns in materials discovery, where samples with high uncertainty are interpreted as containing information missing from the model. In this work, the effect of different uncertainty estimation and calibration methods are evaluated for active learning when using ensembles of ALIGNN, eXtreme Gradient Boost, Random Forest, and Neural Network model architectures. We compare uncertainty estimates from ALIGNN deep ensembles to loss landscape uncertainty estimates obtained for solubility, bandgap, and formation energy prediction tasks. We then evaluate how the quality of the uncertainty estimate impacts an active learning campaign that seeks model generalization to out-of-distribution data. Uncertainty calibration methods were found to variably generalize from in-domain data to out-of-domain data. Furthermore, calibrated uncertainties were generally unsuccessful in reducing the amount of data required by a model to improve during an active learning campaign on out-of-distribution data when compared to random sampling and uncalibrated uncertainties. The impact of poor-quality uncertainty persists for random forest and eXtreme Gradient Boosting models trained on the same data for the same tasks, indicating that this is at least partially intrinsic to the data and not due to model capacity alone. Analysis of the target, in-distribution uncertainty, out-of-distribution uncertainty, and training residual distributions suggest that future work focus on understanding empirical uncertainties in the feature input space for cases where ensemble prediction variances do not accurately capture the missing information required for the model to generalize.", "date": "2025-11-21", "categories": []}
{"id": "2511.18007", "title": "Is Complete Labeling Necessary? Understanding Active Learning in Longitudinal Medical Imaging", "abstract": "Detecting changes in longitudinal medical imaging using deep learning requires a substantial amount of accurately labeled data. However, labeling these images is notably more costly and time-consuming than labeling other image types, as it requires labeling across various time points, where new lesions can be minor, and subtle changes are easily missed. Deep Active Learning (DAL) has shown promise in minimizing labeling costs by selectively querying the most informative samples, but existing studies have primarily focused on static tasks like classification and segmentation. Consequently, the conventional DAL approach cannot be directly applied to change detection tasks, which involve identifying subtle differences across multiple images. In this study, we propose a novel DAL framework, named Longitudinal Medical Imaging Active Learning (LMI-AL), tailored specifically for longitudinal medical imaging. By pairing and differencing all 2D slices from baseline and follow-up 3D images, LMI-AL iteratively selects the most informative pairs for labeling using DAL, training a deep learning model with minimal manual annotation. Experimental results demonstrate that, with less than 8% of the data labeled, LMI-AL can achieve performance comparable to models trained on fully labeled datasets. We also provide a detailed analysis of the method's performance, as guidance for future research. The code is publicly available at https://github.com/HelenMa9998/Longitudinal_AL.", "date": "2025-11-22", "categories": []}
{"id": "2511.19183", "title": "nnActive: A Framework for Evaluation of Active Learning in 3D Biomedical Segmentation", "abstract": "Semantic segmentation is crucial for various biomedical applications, yet its reliance on large annotated datasets presents a bottleneck due to the high cost and specialized expertise required for manual labeling. Active Learning (AL) aims to mitigate this challenge by querying only the most informative samples, thereby reducing annotation effort. However, in the domain of 3D biomedical imaging, there is no consensus on whether AL consistently outperforms Random sampling. Four evaluation pitfalls hinder the current methodological assessment. These are (1) restriction to too few datasets and annotation budgets, (2) using 2D models on 3D images without partial annotations, (3) Random baseline not being adapted to the task, and (4) measuring annotation cost only in voxels. In this work, we introduce nnActive, an open-source AL framework that overcomes these pitfalls by (1) means of a large scale study spanning four biomedical imaging datasets and three label regimes, (2) extending nnU-Net by using partial annotations for training with 3D patch-based query selection, (3) proposing Foreground Aware Random sampling strategies tackling the foreground-background class imbalance of medical images and (4) propose the foreground efficiency metric, which captures the low annotation cost of background-regions. We reveal the following findings: (A) while all AL methods outperform standard Random sampling, none reliably surpasses an improved Foreground Aware Random sampling; (B) benefits of AL depend on task specific parameters; (C) Predictive Entropy is overall the best performing AL method, but likely requires the most annotation effort; (D) AL performance can be improved with more compute intensive design choices. As a holistic, open-source framework, nnActive can serve as a catalyst for research and application of AL in 3D biomedical imaging. Code is at: https://github.com/MIC-DKFZ/nnActive", "date": "2025-11-24", "categories": []}
{"id": "2511.19301", "title": "IDEAL-M3D: Instance Diversity-Enriched Active Learning for Monocular 3D Detection", "abstract": "Monocular 3D detection relies on just a single camera and is therefore easy to deploy. Yet, achieving reliable 3D understanding from monocular images requires substantial annotation, and 3D labels are especially costly. To maximize performance under constrained labeling budgets, it is essential to prioritize annotating samples expected to deliver the largest performance gains. This prioritization is the focus of active learning. Curiously, we observed two significant limitations in active learning algorithms for 3D monocular object detection. First, previous approaches select entire images, which is inefficient, as non-informative instances contained in the same image also need to be labeled. Secondly, existing methods rely on uncertainty-based selection, which in monocular 3D object detection creates a bias toward depth ambiguity. Consequently, distant objects are selected, while nearby objects are overlooked.\n  To address these limitations, we propose IDEAL-M3D, the first instance-level pipeline for monocular 3D detection. For the first time, we demonstrate that an explicitly diverse, fast-to-train ensemble improves diversity-driven active learning for monocular 3D. We induce diversity with heterogeneous backbones and task-agnostic features, loss weight perturbation, and time-dependent bagging. IDEAL-M3D shows superior performance and significant resource savings: with just 60% of the annotations, we achieve similar or better AP3D on KITTI validation and test set results compared to training the same detector on the whole dataset.", "date": "2025-11-24", "categories": []}
{"id": "2505.01160", "title": "TActiLE: Tiny Active LEarning for wearable devices", "abstract": "Tiny Machine Learning (TinyML) algorithms have seen extensive use in recent years, enabling wearable devices to be not only connected but also genuinely intelligent by running machine learning (ML) computations directly on-device. Among such devices, smart glasses have particularly benefited from TinyML advancements. TinyML facilitates the on-device execution of the inference phase of ML algorithms on embedded and wearable devices, and more recently, it has expanded into On-device Learning (ODL), which allows both inference and learning phases to occur directly on the device. The application of ODL techniques to wearable devices is particularly compelling, as it enables the development of more personalized models that adapt based on the data of the user. However, one of the major challenges of ODL algorithms is the scarcity of labeled data collected on-device. In smart wearable contexts, requiring users to manually label large amounts of data is often impractical and could lead to user disengagement with the technology. To address this issue, this paper explores the application of Active Learning (AL) techniques, i.e., techniques that aim at minimizing the labeling effort, by actively selecting from a large quantity of unlabeled data only a small subset to be labeled and added to the training set of the algorithm. In particular, we propose TActiLE, a novel AL algorithm that selects from the stream of on-device sensor data the ones that would help the ML algorithm improve the most once coupled with labels provided by the user. TActiLE is the first Active Learning technique specifically designed for the TinyML context. We evaluate its effectiveness and efficiency through experiments on multiple image classification datasets. The results demonstrate its suitability for tiny and wearable devices.", "date": "2025-05-02", "categories": []}
{"id": "2507.01372", "title": "Active Measurement: Efficient Estimation at Scale", "abstract": "AI has the potential to transform scientific discovery by analyzing vast datasets with little human effort. However, current workflows often do not provide the accuracy or statistical guarantees that are needed. We introduce active measurement, a human-in-the-loop AI framework for scientific measurement. An AI model is used to predict measurements for individual units, which are then sampled for human labeling using importance sampling. With each new set of human labels, the AI model is improved and an unbiased Monte Carlo estimate of the total measurement is refined. Active measurement can provide precise estimates even with an imperfect AI model, and requires little human effort when the AI model is very accurate. We derive novel estimators, weighting schemes, and confidence intervals, and show that active measurement reduces estimation error compared to alternatives in several measurement tasks.", "date": "2025-07-02", "categories": []}
{"id": "2505.19404", "title": "Exploring the Possibility of TypiClust for Low-Budget Federated Active Learning", "abstract": "Federated Active Learning (FAL) seeks to reduce the burden of annotation under the realistic constraints of federated learning by leveraging Active Learning (AL). As FAL settings make it more expensive to obtain ground truth labels, FAL strategies that work well in low-budget regimes, where the amount of annotation is very limited, are needed. In this work, we investigate the effectiveness of TypiClust, a successful low-budget AL strategy, in low-budget FAL settings. Our empirical results show that TypiClust works well even in low-budget FAL settings contrasted with relatively low performances of other methods, although these settings present additional challenges, such as data heterogeneity, compared to AL. In addition, we show that FAL settings cause distribution shifts in terms of typicality, but TypiClust is not very vulnerable to the shifts. We also analyze the sensitivity of TypiClust to feature extraction methods, and it suggests a way to perform FAL even in limited data situations.", "date": "2025-05-26", "categories": []}
{"id": "2511.15132", "title": "WaveFuse-AL: Cyclical and Performance-Adaptive Multi-Strategy Active Learning for Medical Images", "abstract": "Active learning reduces annotation costs in medical imaging by strategically selecting the most informative samples for labeling. However, individual acquisition strategies often exhibit inconsistent behavior across different stages of the active learning cycle. We propose Cyclical and Performance-Adaptive Multi-Strategy Active Learning (WaveFuse-AL), a novel framework that adaptively fuses multiple established acquisition strategies-BALD, BADGE, Entropy, and CoreSet throughout the learning process. WaveFuse-AL integrates cyclical (sinusoidal) temporal priors with performance-driven adaptation to dynamically adjust strategy importance over time. We evaluate WaveFuse-AL on three medical imaging benchmarks: APTOS-2019 (multi-class classification), RSNA Pneumonia Detection (binary classification), and ISIC-2018 (skin lesion segmentation). Experimental results demonstrate that WaveFuse-AL consistently outperforms both single-strategy and alternating-strategy baselines, achieving statistically significant performance improvements (on ten out of twelve metric measurements) while maximizing the utility of limited annotation budgets.", "date": "2025-11-19", "categories": []}
{"id": "2511.14028", "title": "LINGUAL: Language-INtegrated GUidance in Active Learning for Medical Image Segmentation", "abstract": "Although active learning (AL) in segmentation tasks enables experts to annotate selected regions of interest (ROIs) instead of entire images, it remains highly challenging, labor-intensive, and cognitively demanding due to the blurry and ambiguous boundaries commonly observed in medical images. Also, in conventional AL, annotation effort is a function of the ROI- larger regions make the task cognitively easier but incur higher annotation costs, whereas smaller regions demand finer precision and more attention from the expert. In this context, language guidance provides an effective alternative, requiring minimal expert effort while bypassing the cognitively demanding task of precise boundary delineation in segmentation. Towards this goal, we introduce LINGUAL: a framework that receives natural language instructions from an expert, translates them into executable programs through in-context learning, and automatically performs the corresponding sequence of sub-tasks without any human intervention. We demonstrate the effectiveness of LINGUAL in active domain adaptation (ADA) achieving comparable or superior performance to AL baselines while reducing estimated annotation time by approximately 80%.", "date": "2025-11-18", "categories": []}
{"id": "2509.08988", "title": "Active Learning and Explainable AI for Multi-Objective Optimization of Spin Coated Polymers", "abstract": "Spin coating polymer thin films to achieve specific mechanical properties is inherently a multi-objective optimization problem. We present a framework that integrates an active Pareto front learning algorithm (PyePAL) with visualization and explainable AI techniques to optimize processing parameters. PyePAL uses Gaussian process models to predict objective values (hardness and elasticity) from the design variables (spin speed, dilution, and polymer mixture), guiding the adaptive selection of samples toward promising regions of the design space. To enable interpretable insights into the high-dimensional design space, we utilize UMAP (Uniform Manifold Approximation and Projection) for two-dimensional visualization of the Pareto front exploration. Additionally, we incorporate fuzzy linguistic summaries, which translate the learned relationships between process parameters and performance objectives into linguistic statements, thus enhancing the explainability and understanding of the optimization results. Experimental results demonstrate that our method efficiently identifies promising polymer designs, while the visual and linguistic explanations facilitate expert-driven analysis and knowledge discovery.", "date": "2025-09-10", "categories": []}
{"id": "2401.16982", "title": "ActDroid: An active learning framework for Android malware detection", "abstract": "The growing popularity of Android requires malware detection systems that can keep up with the pace of new software being released. According to a recent study, a new piece of malware appears online every 12 seconds. To address this, we treat Android malware detection as a streaming data problem and explore the use of active online learning as a means of mitigating the problem of labelling applications in a timely and cost-effective manner. Our resulting framework achieves accuracies of up to 96\\%, requires as little of 24\\% of the training data to be labelled, and compensates for concept drift that occurs between the release and labelling of an application. We also consider the broader practicalities of online learning within Android malware detection, and systematically explore the trade-offs between using different static, dynamic and hybrid feature sets to classify malware.", "date": "2024-01-30", "categories": []}
{"id": "2511.08991", "title": "Robust Sampling for Active Statistical Inference", "abstract": "Active statistical inference is a new method for inference with AI-assisted data collection. Given a budget on the number of labeled data points that can be collected and assuming access to an AI predictive model, the basic idea is to improve estimation accuracy by prioritizing the collection of labels where the model is most uncertain. The drawback, however, is that inaccurate uncertainty estimates can make active sampling produce highly noisy results, potentially worse than those from naive uniform sampling. In this work, we present robust sampling strategies for active statistical inference. Robust sampling ensures that the resulting estimator is never worse than the estimator using uniform sampling. Furthermore, with reliable uncertainty estimates, the estimator usually outperforms standard active inference. This is achieved by optimally interpolating between uniform and active sampling, depending on the quality of the uncertainty scores, and by using ideas from robust optimization. We demonstrate the utility of the method on a series of real datasets from computational social science and survey research.", "date": "2025-11-12", "categories": []}
{"id": "2511.06658", "title": "Active Learning for Animal Re-Identification with Ambiguity-Aware Sampling", "abstract": "Animal Re-ID has recently gained substantial attention in the AI research community due to its high impact on biodiversity monitoring and unique research challenges arising from environmental factors. The subtle distinguishing patterns, handling new species and the inherent open-set nature make the problem even harder. To address these complexities, foundation models trained on labeled, large-scale and multi-species animal Re-ID datasets have recently been introduced to enable zero-shot Re-ID. However, our benchmarking reveals significant gaps in their zero-shot Re-ID performance for both known and unknown species. While this highlights the need for collecting labeled data in new domains, exhaustive annotation for Re-ID is laborious and requires domain expertise. Our analyses show that existing unsupervised (USL) and AL Re-ID methods underperform for animal Re-ID. To address these limitations, we introduce a novel AL Re-ID framework that leverages complementary clustering methods to uncover and target structurally ambiguous regions in the embedding space for mining pairs of samples that are both informative and broadly representative. Oracle feedback on these pairs, in the form of must-link and cannot-link constraints, facilitates a simple annotation interface, which naturally integrates with existing USL methods through our proposed constrained clustering refinement algorithm. Through extensive experiments, we demonstrate that, by utilizing only 0.033% of all annotations, our approach consistently outperforms existing foundational, USL and AL baselines. Specifically, we report an average improvement of 10.49%, 11.19% and 3.99% (mAP) on 13 wildlife datasets over foundational, USL and AL methods, respectively, while attaining state-of-the-art performance on each dataset. Furthermore, we also show an improvement of 11.09%, 8.2% and 2.06% for unknown individuals in an open-world setting.", "date": "2025-11-10", "categories": []}
{"id": "2511.05736", "title": "Near-Exponential Savings for Mean Estimation with Active Learning", "abstract": "We study the problem of efficiently estimating the mean of a $k$-class random variable, $Y$, using a limited number of labels, $N$, in settings where the analyst has access to auxiliary information (i.e.: covariates) $X$ that may be informative about $Y$. We propose an active learning algorithm (\"PartiBandits\") to estimate $\\mathbb{E}[Y]$. The algorithm yields an estimate, $\\widehat{\\mu}_{\\text{PB}}$, such that $\\left( \\widehat{\\mu}_{\\text{PB}} - \\mathbb{E}[Y]\\right)^2$ is $\\tilde{\\mathcal{O}}\\left( \\frac{\\nu + \\exp(c \\cdot (-N/\\log(N))) }{N} \\right)$, where $c > 0$ is a constant and $\\nu$ is the risk of the Bayes-optimal classifier. PartiBandits is essentially a two-stage algorithm. In the first stage, it learns a partition of the unlabeled data that shrinks the average conditional variance of $Y$. In the second stage it uses a UCB-style subroutine (\"WarmStart-UCB\") to request labels from each stratum round-by-round. Both the main algorithm's and the subroutine's convergence rates are minimax optimal in classical settings. PartiBandits bridges the UCB and disagreement-based approaches to active learning despite these two approaches being designed to tackle very different tasks. We illustrate our methods through simulation using nationwide electronic health records. Our methods can be implemented using the PartiBandits package in R.", "date": "2025-11-07", "categories": []}
{"id": "2508.05090", "title": "Cold-Start Active Preference Learning in Socio-Economic Domains", "abstract": "Active preference learning offers an efficient approach to modeling preferences, but it is hindered by the cold-start problem, which leads to a marked decline in performance when no initial labeled data are available. While cold-start solutions have been proposed for domains such as vision and text, the cold-start problem in active preference learning remains largely unexplored, underscoring the need for practical, effective methods. Drawing inspiration from established practices in social and economic research, the proposed method initiates learning with a self-supervised phase that employs Principal Component Analysis (PCA) to generate initial pseudo-labels. This process produces a \\say{warmed-up} model based solely on the data's intrinsic structure, without requiring expert input. The model is then refined through an active learning loop that strategically queries a simulated noisy oracle for labels. Experiments conducted on various socio-economic datasets, including those related to financial credibility, career success rate, and socio-economic status, consistently show that the PCA-driven approach outperforms standard active learning strategies that start without prior information. This work thus provides a computationally efficient and straightforward solution that effectively addresses the cold-start problem.", "date": "2025-08-07", "categories": []}
{"id": "2510.24053", "title": "Low-N Protein Activity Optimization with FolDE", "abstract": "Proteins are traditionally optimized through the costly construction and measurement of many mutants. Active Learning-assisted Directed Evolution (ALDE) alleviates that cost by predicting the best improvements and iteratively testing mutants to inform predictions. However, existing ALDE methods face a critical limitation: selecting the highest-predicted mutants in each round yields homogeneous training data insufficient for accurate prediction models in subsequent rounds. Here we present FolDE, an ALDE method designed to maximize end-of-campaign success. In simulations across 20 protein targets, FolDE discovers 23% more top 10% mutants than the best baseline ALDE method (p=0.005) and is 55% more likely to find top 1% mutants. FolDE achieves this primarily through naturalness-based warm-starting, which augments limited activity measurements with protein language model outputs to improve activity prediction. We also introduce a constant-liar batch selector, which improves batch diversity; this is important in multi-mutation campaigns but had limited effect in our benchmarks. The complete workflow is freely available as open-source software, making efficient protein optimization accessible to any laboratory.", "date": "2025-10-28", "categories": []}
{"id": "2505.22494", "title": "ProSpero: Active Learning for Robust Protein Design Beyond Wild-Type Neighborhoods", "abstract": "Designing protein sequences of both high fitness and novelty is a challenging task in data-efficient protein engineering. Exploration beyond wild-type neighborhoods often leads to biologically implausible sequences or relies on surrogate models that lose fidelity in novel regions. Here, we propose ProSpero, an active learning framework in which a frozen pre-trained generative model is guided by a surrogate updated from oracle feedback. By integrating fitness-relevant residue selection with biologically-constrained Sequential Monte Carlo sampling, our approach enables exploration beyond wild-type neighborhoods while preserving biological plausibility. We show that our framework remains effective even when the surrogate is misspecified. ProSpero consistently outperforms or matches existing methods across diverse protein engineering tasks, retrieving sequences of both high fitness and novelty.", "date": "2025-05-28", "categories": []}
{"id": "2510.22214", "title": "GALA: A GlobAl-LocAl Approach for Multi-Source Active Domain Adaptation", "abstract": "Domain Adaptation (DA) provides an effective way to tackle target-domain tasks by leveraging knowledge learned from source domains. Recent studies have extended this paradigm to Multi-Source Domain Adaptation (MSDA), which exploits multiple source domains carrying richer and more diverse transferable information. However, a substantial performance gap still remains between adaptation-based methods and fully supervised learning. In this paper, we explore a more practical and challenging setting, named Multi-Source Active Domain Adaptation (MS-ADA), to further enhance target-domain performance by selectively acquiring annotations from the target domain. The key difficulty of MS-ADA lies in designing selection criteria that can jointly handle inter-class diversity and multi-source domain variation. To address these challenges, we propose a simple yet effective GALA strategy (GALA), which combines a global k-means clustering step for target-domain samples with a cluster-wise local selection criterion, effectively tackling the above two issues in a complementary manner. Our proposed GALA is plug-and-play and can be seamlessly integrated into existing DA frameworks without introducing any additional trainable parameters. Extensive experiments on three standard DA benchmarks demonstrate that GALA consistently outperforms prior active learning and active DA methods, achieving performance comparable to the fully-supervised upperbound while using only 1% of the target annotations.", "date": "2025-10-25", "categories": []}
{"id": "2510.22229", "title": "Diffusion-Driven Two-Stage Active Learning for Low-Budget Semantic Segmentation", "abstract": "Semantic segmentation demands dense pixel-level annotations, which can be prohibitively expensive - especially under extremely constrained labeling budgets. In this paper, we address the problem of low-budget active learning for semantic segmentation by proposing a novel two-stage selection pipeline. Our approach leverages a pre-trained diffusion model to extract rich multi-scale features that capture both global structure and fine details. In the first stage, we perform a hierarchical, representation-based candidate selection by first choosing a small subset of representative pixels per image using MaxHerding, and then refining these into a diverse global pool. In the second stage, we compute an entropy-augmented disagreement score (eDALD) over noisy multi-scale diffusion features to capture both epistemic uncertainty and prediction confidence, selecting the most informative pixels for annotation. This decoupling of diversity and uncertainty lets us achieve high segmentation accuracy with only a tiny fraction of labeled pixels. Extensive experiments on four benchmarks (CamVid, ADE-Bed, Cityscapes, and Pascal-Context) demonstrate that our method significantly outperforms existing baselines under extreme pixel-budget regimes. Our code is available at https://github.com/jn-kim/two-stage-edald.", "date": "2025-10-25", "categories": []}
{"id": "2510.16757", "title": "SAMOSA: Sharpness Aware Minimization for Open Set Active learning", "abstract": "Modern machine learning solutions require extensive data collection where labeling remains costly. To reduce this burden, open set active learning approaches aim to select informative samples from a large pool of unlabeled data that includes irrelevant or unknown classes. In this context, we propose Sharpness Aware Minimization for Open Set Active Learning (SAMOSA) as an effective querying algorithm. Building on theoretical findings concerning the impact of data typicality on the generalization properties of traditional stochastic gradient descent (SGD) and sharpness-aware minimization (SAM), SAMOSA actively queries samples based on their typicality. SAMOSA effectively identifies atypical samples that belong to regions of the embedding manifold close to the model decision boundaries. Therefore, SAMOSA prioritizes the samples that are (i) highly informative for the targeted classes, and (ii) useful for distinguishing between targeted and unwanted classes. Extensive experiments show that SAMOSA achieves up to 3% accuracy improvement over the state of the art across several datasets, while not introducing computational overhead. The source code of our experiments is available at: https://anonymous.4open.science/r/samosa-DAF4", "date": "2025-10-19", "categories": []}
{"id": "2506.07259", "title": "ALINE: Joint Amortization for Bayesian Inference and Active Data Acquisition", "abstract": "Many critical applications, from autonomous scientific discovery to personalized medicine, demand systems that can both strategically acquire the most informative data and instantaneously perform inference based upon it. While amortized methods for Bayesian inference and experimental design offer part of the solution, neither approach is optimal in the most general and challenging task, where new data needs to be collected for instant inference. To tackle this issue, we introduce the Amortized Active Learning and Inference Engine (ALINE), a unified framework for amortized Bayesian inference and active data acquisition. ALINE leverages a transformer architecture trained via reinforcement learning with a reward based on self-estimated information gain provided by its own integrated inference component. This allows it to strategically query informative data points while simultaneously refining its predictions. Moreover, ALINE can selectively direct its querying strategy towards specific subsets of model parameters or designated predictive tasks, optimizing for posterior estimation, data prediction, or a mixture thereof. Empirical results on regression-based active learning, classical Bayesian experimental design benchmarks, and a psychometric model with selectively targeted parameters demonstrate that ALINE delivers both instant and accurate inference along with efficient selection of informative points.", "date": "2025-06-08", "categories": []}
{"id": "2510.18660", "title": "Image augmentation with invertible networks in interactive satellite image change detection", "abstract": "This paper devises a novel interactive satellite image change detection algorithm based on active learning. Our framework employs an iterative process that leverages a question-and-answer model. This model queries the oracle (user) about the labels of a small subset of images (dubbed as display), and based on the oracle's responses, change detection model is dynamically updated. The main contribution of our framework resides in a novel invertible network that allows augmenting displays, by mapping them from highly nonlinear input spaces to latent ones, where augmentation transformations become linear and more tractable. The resulting augmented data are afterwards mapped back to the input space, and used to retrain more effective change detection criteria in the subsequent iterations of active learning. Experimental results demonstrate superior performance of our proposed method compared to the related work.", "date": "2025-10-21", "categories": []}
{"id": "2501.11039", "title": "Model Predictive Task Sampling for Efficient and Robust Adaptation", "abstract": "Foundation models have revolutionized general-purpose problem-solving, offering rapid task adaptation through pretraining, meta-training, and finetuning. Recent crucial advances in these paradigms reveal the importance of challenging task prioritized sampling to enhance adaptation robustness under distribution shifts. However, ranking task difficulties over iteration as a preliminary step typically requires exhaustive task evaluation, which is practically unaffordable in computation and data-annotation. This study provides a novel perspective to illuminate the possibility of leveraging the dual importance of adaptation robustness and learning efficiency, particularly in scenarios where task evaluation is risky or costly, such as iterative agent-environment interactions for robotic policy evaluation or computationally intensive inference steps for finetuning foundation models. Firstly, we introduce Model Predictive Task Sampling (MPTS), a framework that bridges the task space and adaptation risk distributions, providing a theoretical foundation for robust active task sampling. MPTS employs a generative model to characterize the episodic optimization process and predicts task-specific adaptation risk via posterior inference. The resulting risk predictive model amortizes the costly evaluation of task adaptation performance and provably approximates task difficulty rankings. MPTS seamlessly integrates into zero-shot, few-shot, and supervised finetuning settings. Empirically, we conduct extensive experiments in pattern recognition using foundation models and sequential decision-making. Our results demonstrate that MPTS significantly enhances adaptation robustness for tail risk or out-of-distribution (OOD) tasks and improves learning efficiency compared to state-of-the-art (SoTA) methods. The code is available at the project site https://github.com/thu-rllab/MPTS.", "date": "2025-01-19", "categories": []}
{"id": "2505.06535", "title": "Online Feedback Efficient Active Target Discovery in Partially Observable Environments", "abstract": "In various scientific and engineering domains, where data acquisition is costly--such as in medical imaging, environmental monitoring, or remote sensing--strategic sampling from unobserved regions, guided by prior observations, is essential to maximize target discovery within a limited sampling budget. In this work, we introduce Diffusion-guided Active Target Discovery (DiffATD), a novel method that leverages diffusion dynamics for active target discovery. DiffATD maintains a belief distribution over each unobserved state in the environment, using this distribution to dynamically balance exploration-exploitation. Exploration reduces uncertainty by sampling regions with the highest expected entropy, while exploitation targets areas with the highest likelihood of discovering the target, indicated by the belief distribution and an incrementally trained reward model designed to learn the characteristics of the target. DiffATD enables efficient target discovery in a partially observable environment within a fixed sampling budget, all without relying on any prior supervised training. Furthermore, DiffATD offers interpretability, unlike existing black--box policies that require extensive supervised training. Through extensive experiments and ablation studies across diverse domains, including medical imaging, species discovery, and remote sensing, we show that DiffATD performs significantly better than baselines and competitively with supervised methods that operate under full environmental observability.", "date": "2025-05-10", "categories": []}
{"id": "2510.16676", "title": "Active Target Discovery under Uninformative Prior: The Power of Permanent and Transient Memory", "abstract": "In many scientific and engineering fields, where acquiring high-quality data is expensive--such as medical imaging, environmental monitoring, and remote sensing--strategic sampling of unobserved regions based on prior observations is crucial for maximizing discovery rates within a constrained budget. The rise of powerful generative models, such as diffusion models, has enabled active target discovery in partially observable environments by leveraging learned priors--probabilistic representations that capture underlying structure from data. With guidance from sequentially gathered task-specific observations, these models can progressively refine exploration and efficiently direct queries toward promising regions. However, in domains where learning a strong prior is infeasible due to extremely limited data or high sampling cost (such as rare species discovery, diagnostics for emerging diseases, etc.), these methods struggle to generalize. To overcome this limitation, we propose a novel approach that enables effective active target discovery even in settings with uninformative priors, ensuring robust exploration and adaptability in complex real-world scenarios. Our framework is theoretically principled and draws inspiration from neuroscience to guide its design. Unlike black-box policies, our approach is inherently interpretable, providing clear insights into decision-making. Furthermore, it guarantees a strong, monotonic improvement in prior estimates with each new observation, leading to increasingly accurate sampling and reinforcing both reliability and adaptability in dynamic settings. Through comprehensive experiments and ablation studies across various domains, including species distribution modeling and remote sensing, we demonstrate that our method substantially outperforms baseline approaches.", "date": "2025-10-19", "categories": []}
{"id": "2510.12144", "title": "Budget-constrained Active Learning to Effectively De-censor Survival Data", "abstract": "Standard supervised learners attempt to learn a model from a labeled dataset. Given a small set of labeled instances, and a pool of unlabeled instances, a budgeted learner can use its given budget to pay to acquire the labels of some unlabeled instances, which it can then use to produce a model. Here, we explore budgeted learning in the context of survival datasets, which include (right) censored instances, where we know only a lower bound on an instance's time-to-event. Here, that learner can pay to (partially) label a censored instance -- e.g., to acquire the actual time for an instance [perhaps go from (3 yr, censored) to (7.2 yr, uncensored)], or other variants [e.g., learn about one more year, so go from (3 yr, censored) to either (4 yr, censored) or perhaps (3.2 yr, uncensored)]. This serves as a model of real world data collection, where follow-up with censored patients does not always lead to uncensoring, and how much information is given to the learner model during data collection is a function of the budget and the nature of the data itself. We provide both experimental and theoretical results for how to apply state-of-the-art budgeted learning algorithms to survival data and the respective limitations that exist in doing so. Our approach provides bounds and time complexity asymptotically equivalent to the standard active learning method BatchBALD. Moreover, empirical analysis on several survival tasks show that our model performs better than other potential approaches on several benchmarks.", "date": "2025-10-14", "categories": []}
{"id": "2510.09740", "title": "Reliable Active Learning from Unreliable Labels via Neural Collapse Geometry", "abstract": "Active Learning (AL) promises to reduce annotation cost by prioritizing informative samples, yet its reliability is undermined when labels are noisy or when the data distribution shifts. In practice, annotators make mistakes, rare categories are ambiguous, and conventional AL heuristics (uncertainty, diversity) often amplify such errors by repeatedly selecting mislabeled or redundant samples. We propose Reliable Active Learning via Neural Collapse Geometry (NCAL-R), a framework that leverages the emergent geometric regularities of deep networks to counteract unreliable supervision. Our method introduces two complementary signals: (i) a Class-Mean Alignment Perturbation score, which quantifies how candidate samples structurally stabilize or distort inter-class geometry, and (ii) a Feature Fluctuation score, which captures temporal instability of representations across training checkpoints. By combining these signals, NCAL-R prioritizes samples that both preserve class separation and highlight ambiguous regions, mitigating the effect of noisy or redundant labels. Experiments on ImageNet-100 and CIFAR100 show that NCAL-R consistently outperforms standard AL baselines, achieving higher accuracy with fewer labels, improved robustness under synthetic label noise, and stronger generalization to out-of-distribution data. These results suggest that incorporating geometric reliability criteria into acquisition decisions can make Active Learning less brittle to annotation errors and distribution shifts, a key step toward trustworthy deployment in real-world labeling pipelines. Our code is available at https://github.com/Vision-IIITD/NCAL.", "date": "2025-10-10", "categories": []}
{"id": "2510.08865", "title": "Multi-fidelity Batch Active Learning for Gaussian Process Classifiers", "abstract": "Many science and engineering problems rely on expensive computational simulations, where a multi-fidelity approach can accelerate the exploration of a parameter space. We study efficient allocation of a simulation budget using a Gaussian Process (GP) model in the binary simulation output case. This paper introduces Bernoulli Parameter Mutual Information (BPMI), a batch active learning algorithm for multi-fidelity GP classifiers. BPMI circumvents the intractability of calculating mutual information in the probability space by employing a first-order Taylor expansion of the link function. We evaluate BPMI against several baselines on two synthetic test cases and a complex, real-world application involving the simulation of a laser-ignited rocket combustor. In all experiments, BPMI demonstrates superior performance, achieving higher predictive accuracy for a fixed computational budget.", "date": "2025-10-09", "categories": []}
{"id": "2510.09418", "title": "Active Model Selection for Large Language Models", "abstract": "We introduce LLM SELECTOR, the first framework for active model selection of Large Language Models (LLMs). Unlike prior evaluation and benchmarking approaches that rely on fully annotated datasets, LLM SELECTOR efficiently identifies the best LLM with limited annotations. In particular, for any given task, LLM SELECTOR adaptively selects a small set of queries to annotate that are most informative about the best model for the task. To further reduce annotation cost, we leverage a judge-based oracle annotation model. Through extensive experiments on 6 benchmarks with 151 LLMs, we show that LLM SELECTOR reduces annotation costs by up to 59.62% when selecting the best and near-best LLM for the task.", "date": "2025-10-10", "categories": []}
{"id": "2503.05969", "title": "Dependency-aware Maximum Likelihood Estimation for Active Learning", "abstract": "Active learning aims to efficiently build a labeled training set by strategically selecting samples to query labels from annotators. In this sequential process, each sample acquisition influences subsequent selections, causing dependencies among samples in the labeled set. However, these dependencies are overlooked during the model parameter estimation stage when updating the model using Maximum Likelihood Estimation (MLE), a conventional method that assumes independent and identically distributed (i.i.d.) data. We propose Dependency-aware MLE (DMLE), which corrects MLE within the active learning framework by addressing sample dependencies typically neglected due to the i.i.d. assumption, ensuring consistency with active learning principles in the model parameter estimation process. This improved method achieves superior performance across multiple benchmark datasets, reaching higher performance in earlier cycles compared to conventional MLE. Specifically, we observe average accuracy improvements of 6%, 8.6%, and 10.5% for k=1, k=5, and k=10 respectively, after collecting the first 100 samples, where entropy is the acquisition function and k is the query batch size acquired at every active learning cycle. Our implementation is publicly available at: https://github.com/neu-spiral/DMLEforAL", "date": "2025-03-07", "categories": []}
{"id": "2510.04939", "title": "Unsupervised Active Learning via Natural Feature Progressive Framework", "abstract": "The effectiveness of modern deep learning models is predicated on the availability of large-scale, human-annotated datasets, a process that is notoriously expensive and time-consuming. While Active Learning (AL) offers a strategic solution by labeling only the most informative and representative data, its iterative nature still necessitates significant human involvement. Unsupervised Active Learning (UAL) presents an alternative by shifting the annotation burden to a single, post-selection step. Unfortunately, prevailing UAL methods struggle to achieve state-of-the-art performance. These approaches typically rely on local, gradient-based scoring for sample importance estimation, which not only makes them vulnerable to ambiguous and noisy data but also hinders their capacity to select samples that adequately represent the full data distribution. Moreover, their use of shallow, one-shot linear selection falls short of a true UAL paradigm. In this paper, we propose the Natural Feature Progressive Framework (NFPF), a UAL method that revolutionizes how sample importance is measured. At its core, NFPF employs a Specific Feature Learning Machine (SFLM) to effectively quantify each sample's contribution to model performance. We further utilize the SFLM to define a powerful Reconstruction Difference metric for initial sample selection. Our comprehensive experiments show that NFPF significantly outperforms all established UAL methods and achieves performance on par with supervised AL methods on vision datasets. Detailed ablation studies and qualitative visualizations provide compelling evidence for NFPF's superior performance, enhanced robustness, and improved data distribution coverage.", "date": "2025-10-06", "categories": []}
{"id": "2510.03162", "title": "Calibrated Uncertainty Sampling for Active Learning", "abstract": "We study the problem of actively learning a classifier with a low calibration error. One of the most popular Acquisition Functions (AFs) in pool-based Active Learning (AL) is querying by the model's uncertainty. However, we recognize that an uncalibrated uncertainty model on the unlabeled pool may significantly affect the AF effectiveness, leading to sub-optimal generalization and high calibration error on unseen data. Deep Neural Networks (DNNs) make it even worse as the model uncertainty from DNN is usually uncalibrated. Therefore, we propose a new AF by estimating calibration errors and query samples with the highest calibration error before leveraging DNN uncertainty. Specifically, we utilize a kernel calibration error estimator under the covariate shift and formally show that AL with this AF eventually leads to a bounded calibration error on the unlabeled pool and unseen test data. Empirically, our proposed method surpasses other AF baselines by having a lower calibration and generalization error across pool-based AL settings.", "date": "2025-10-03", "categories": []}
{"id": "2506.00910", "title": "PCoreSet: Effective Active Learning through Knowledge Distillation from Vision-Language Models", "abstract": "Knowledge distillation (KD) is a widely used framework for training compact, task-specific models by transferring the knowledge from teacher models. However, its application to active learning (AL), which aims to minimize annotation costs through iterative sample selection, remains underexplored. This gap stems from the fact that KD typically assumes access to sufficient labeled data, whereas AL operates in data-scarce scenarios where task-specific teacher models are often unavailable. In this paper, we first introduce ActiveKD, a framework that integrates AL with KD by leveraging the zero- and few-shot capabilities of large vision-language models (VLMs). A key aspect of ActiveKD is the structured prediction bias of VLMs-i.e., their predictions form clusters in the probability space. We regard this structure as an inductive bias of the teacher model, capturing generalizable output patterns beneficial to student learning. To exploit this bias, we propose Probabilistic CoreSet (PCoreSet), a selection strategy that maximizes coverage in the probability space rather than the feature space. PCoreSet strategically selects probabilistically diverse unlabeled samples, facilitating more efficient transfer of teacher knowledge under limited annotation budgets. Extensive evaluations on 11 datasets show that ActiveKD consistently improves performance across selection methods (e.g., +29.07% on ImageNet, averaged over methods). Under ActiveKD, PCoreSet ranks first in 64/73 settings (approximately 87.7%) across 5 student and 3 teacher networks, always achieving the best performance except for first 2 AL rounds. Our code is available at https://github.com/erjui/PCoreSet.", "date": "2025-06-01", "categories": []}
{"id": "2509.25538", "title": "Steering an Active Learning Workflow Towards Novel Materials Discovery via Queue Prioritization", "abstract": "Generative AI poses both opportunities and risks for solving inverse design problems in the sciences. Generative tools provide the ability to expand and refine a search space autonomously, but do so at the cost of exploring low-quality regions until sufficiently fine tuned. Here, we propose a queue prioritization algorithm that combines generative modeling and active learning in the context of a distributed workflow for exploring complex design spaces. We find that incorporating an active learning model to prioritize top design candidates can prevent a generative AI workflow from expending resources on nonsensical candidates and halt potential generative model decay. For an existing generative AI workflow for discovering novel molecular structure candidates for carbon capture, our active learning approach significantly increases the number of high-quality candidates identified by the generative model. We find that, out of 1000 novel candidates, our workflow without active learning can generate an average of 281 high-performing candidates, while our proposed prioritization with active learning can generate an average 604 high-performing candidates.", "date": "2025-09-29", "categories": []}
{"id": "2509.25692", "title": "Annotation-Efficient Active Test-Time Adaptation with Conformal Prediction", "abstract": "Active Test-Time Adaptation (ATTA) improves model robustness under domain shift by selectively querying human annotations at deployment, but existing methods use heuristic uncertainty measures and suffer from low data selection efficiency, wasting human annotation budget. We propose Conformal Prediction Active TTA (CPATTA), which first brings principled, coverage-guaranteed uncertainty into ATTA. CPATTA employs smoothed conformal scores with a top-K certainty measure, an online weight-update algorithm driven by pseudo coverage, a domain-shift detector that adapts human supervision, and a staged update scheme balances human-labeled and model-labeled data. Extensive experiments demonstrate that CPATTA consistently outperforms the state-of-the-art ATTA methods by around 5% in accuracy. Our code and datasets are available at https://github.com/tingyushi/CPATTA.", "date": "2025-09-30", "categories": []}
{"id": "2509.26564", "title": "Parametric Neural Amp Modeling with Active Learning", "abstract": "We introduce Panama, an active learning framework to train parametric guitar amp models end-to-end using a combination of an LSTM model and a WaveNet-like architecture. With \\model, one can create a virtual amp by recording samples that are determined through an ensemble-based active learning strategy to minimize the amount of datapoints needed (i.e., amp knob settings). Our strategy uses gradient-based optimization to maximize the disagreement among ensemble models, in order to identify the most informative datapoints. MUSHRA listening tests reveal that, with 75 datapoints, our models are able to match the perceptual quality of NAM, the leading open-source non-parametric amp modeler.", "date": "2025-09-30", "categories": []}
{"id": "2509.23515", "title": "From Human Annotation to Automation: LLM-in-the-Loop Active Learning for Arabic Sentiment Analysis", "abstract": "Natural language processing (NLP), particularly sentiment analysis, plays a vital role in areas like marketing, customer service, and social media monitoring by providing insights into user opinions and emotions. However, progress in Arabic sentiment analysis remains limited due to the lack of large, high-quality labeled datasets. While active learning has proven effective in reducing annotation efforts in other languages, few studies have explored it in Arabic sentiment tasks. Likewise, the use of large language models (LLMs) for assisting annotation and comparing their performance to human labeling is still largely unexplored in the Arabic context. In this paper, we propose an active learning framework for Arabic sentiment analysis designed to reduce annotation costs while maintaining high performance. We evaluate multiple deep learning architectures: Specifically, long short-term memory (LSTM), gated recurrent units (GRU), and recurrent neural networks (RNN), across three benchmark datasets: Hunger Station, AJGT, and MASAC, encompassing both modern standard Arabic and dialectal variations. Additionally, two annotation strategies are compared: Human labeling and LLM-assisted labeling. Five LLMs are evaluated as annotators: GPT-4o, Claude 3 Sonnet, Gemini 2.5 Pro, DeepSeek Chat, and LLaMA 3 70B Instruct. For each dataset, the best-performing LLM was used: GPT-4o for Hunger Station, Claude 3 Sonnet for AJGT, and DeepSeek Chat for MASAC. Our results show that LLM-assisted active learning achieves competitive or superior performance compared to human labeling. For example, on the Hunger Station dataset, the LSTM model achieved 93% accuracy with only 450 labeled samples using GPT-4o-generated labels, while on the MASAC dataset, DeepSeek Chat reached 82% accuracy with 650 labeled samples, matching the accuracy obtained through human labeling.", "date": "2025-09-27", "categories": []}
{"id": "2509.24181", "title": "Combining Discrepancy-Confusion Uncertainty and Calibration Diversity for Active Fine-Grained Image Classification", "abstract": "Active learning (AL) aims to build high-quality labeled datasets by iteratively selecting the most informative samples from an unlabeled pool under limited annotation budgets. However, in fine-grained image classification, assessing this informativeness is especially challenging due to subtle inter-class differences. In this paper, we introduce a novel method, combining discrepancy-confusion uncertainty and calibration diversity for active fine-grained image classification (DECERN), to effectively perceive the distinctiveness between fine-grained images and evaluate the sample value. DECERN introduces a multifaceted informativeness measure that combines discrepancy-confusion uncertainty and calibration diversity. The discrepancy-confusion uncertainty quantifies the category directionality and structural stability of fine-grained unlabeled data during local feature fusion. Subsequently, uncertainty-weighted clustering is performed to diversify the uncertainty samples. Then we calibrate the diversity to maximize the global diversity of the selected sample while maintaining its local representativeness. Extensive experiments conducted on 7 fine-grained image datasets across 26 distinct experimental settings demonstrate that our method achieves superior performance compared to state-of-the-art methods.", "date": "2025-09-29", "categories": []}
{"id": "2509.24896", "title": "DAM: Dual Active Learning with Multimodal Foundation Model for Source-Free Domain Adaptation", "abstract": "Source-free active domain adaptation (SFADA) enhances knowledge transfer from a source model to an unlabeled target domain using limited manual labels selected via active learning. While recent domain adaptation studies have introduced Vision-and-Language (ViL) models to improve pseudo-label quality or feature alignment, they often treat ViL-based and data supervision as separate sources, lacking effective fusion. To overcome this limitation, we propose Dual Active learning with Multimodal (DAM) foundation model, a novel framework that integrates multimodal supervision from a ViL model to complement sparse human annotations, thereby forming a dual supervisory signal. DAM initializes stable ViL-guided targets and employs a bidirectional distillation mechanism to foster mutual knowledge exchange between the target model and the dual supervisions during iterative adaptation. Extensive experiments demonstrate that DAM consistently outperforms existing methods and sets a new state-of-the-art across multiple SFADA benchmarks and active learning strategies.", "date": "2025-09-29", "categories": []}
{"id": "2406.01086", "title": "Diverse Subset Selection via Norm-Based Sampling and Orthogonality", "abstract": "Large annotated datasets are crucial for the success of deep neural networks, but labeling data can be prohibitively expensive in domains such as medical imaging. This work tackles the subset selection problem: selecting a small set of the most informative examples from a large unlabeled pool for annotation. We propose a simple and effective method that combines feature norms, randomization, and orthogonality (via the Gram-Schmidt process) to select diverse and informative samples. Feature norms serve as a proxy for informativeness, while randomization and orthogonalization reduce redundancy and encourage coverage of the feature space. Extensive experiments on image and text benchmarks, including CIFAR-10/100, Tiny ImageNet, ImageNet, OrganAMNIST, and Yelp, show that our method consistently improves subset selection performance, both as a standalone approach and when integrated with existing techniques.", "date": "2024-06-03", "categories": []}
{"id": "2509.10742", "title": "Matched-Pair Experimental Design with Active Learning", "abstract": "Matched-pair experimental designs aim to detect treatment effects by pairing participants and comparing within-pair outcome differences. In many situations, the overall effect size across the entire population is small. Then, the focus naturally shifts to identifying and targeting high treatment-effect regions where the intervention is most effective. This paper proposes a matched-pair experimental design that sequentially and actively enrolls patients in high treatment-effect regions. Importantly, we frame the identification of the target region as a classification problem and propose an active learning framework tailored to matched-pair designs. Our design not only reduces the experimental cost of detecting treatment efficacy, but also ensures that the identified regions enclose the entire high-treatment-effect regions. Our theoretical analysis of the framework's label complexity and experiments in practical scenarios demonstrate the efficiency and advantages of the approach.", "date": "2025-09-12", "categories": []}
{"id": "2509.20484", "title": "Data-Efficient Stream-Based Active Distillation for Scalable Edge Model Deployment", "abstract": "Edge camera-based systems are continuously expanding, facing ever-evolving environments that require regular model updates. In practice, complex teacher models are run on a central server to annotate data, which is then used to train smaller models tailored to the edge devices with limited computational power. This work explores how to select the most useful images for training to maximize model quality while keeping transmission costs low. Our work shows that, for a similar training load (i.e., iterations), a high-confidence stream-based strategy coupled with a diversity-based approach produces a high-quality model with minimal dataset queries.", "date": "2025-09-24", "categories": []}
{"id": "2509.20848", "title": "Actively Learning Halfspaces without Synthetic Data", "abstract": "In the classic point location problem, one is given an arbitrary dataset $X \\subset \\mathbb{R}^d$ of $n$ points with query access to an unknown halfspace $f : \\mathbb{R}^d \\to \\{0,1\\}$, and the goal is to learn the label of every point in $X$. This problem is extremely well-studied and a nearly-optimal $\\widetilde{O}(d \\log n)$ query algorithm is known due to Hopkins-Kane-Lovett-Mahajan (FOCS 2020). However, their algorithm is granted the power to query arbitrary points outside of $X$ (point synthesis), and in fact without this power there is an $\\Omega(n)$ query lower bound due to Dasgupta (NeurIPS 2004).\n  In this work our goal is to design efficient algorithms for learning halfspaces without point synthesis. To circumvent the $\\Omega(n)$ lower bound, we consider learning halfspaces whose normal vectors come from a set of size $D$, and show tight bounds of $\\Theta(D + \\log n)$. As a corollary, we obtain an optimal $O(d + \\log n)$ query deterministic learner for axis-aligned halfspaces, closing a previous gap of $O(d \\log n)$ vs. $\\Omega(d + \\log n)$. In fact, our algorithm solves the more general problem of learning a Boolean function $f$ over $n$ elements which is monotone under at least one of $D$ provided orderings. Our technical insight is to exploit the structure in these orderings to perform a binary search in parallel rather than considering each ordering sequentially, and we believe our approach may be of broader interest.\n  Furthermore, we use our exact learning algorithm to obtain nearly optimal algorithms for PAC-learning. We show that $O(\\min(D + \\log(1/\\varepsilon), 1/\\varepsilon) \\cdot \\log D)$ queries suffice to learn $f$ within error $\\varepsilon$, even in a setting when $f$ can be adversarially corrupted on a $c\\varepsilon$-fraction of points, for a sufficiently small constant $c$. This bound is optimal up to a $\\log D$ factor, including in the realizable setting.", "date": "2025-09-25", "categories": []}
{"id": "2509.20003", "title": "Table Detection with Active Learning", "abstract": "Efficient data annotation remains a critical challenge in machine learning, particularly for object detection tasks requiring extensive labeled data. Active learning (AL) has emerged as a promising solution to minimize annotation costs by selecting the most informative samples. While traditional AL approaches primarily rely on uncertainty-based selection, recent advances suggest that incorporating diversity-based strategies can enhance sampling efficiency in object detection tasks. Our approach ensures the selection of representative examples that improve model generalization. We evaluate our method on two benchmark datasets (TableBank-LaTeX, TableBank-Word) using state-of-the-art table detection architectures, CascadeTabNet and YOLOv9. Our results demonstrate that AL-based example selection significantly outperforms random sampling, reducing annotation effort given a limited budget while maintaining comparable performance to fully supervised models. Our method achieves higher mAP scores within the same annotation budget.", "date": "2025-09-24", "categories": []}
{"id": "2407.01804", "title": "DCoM: Active Learning for All Learners", "abstract": "Deep Active Learning (AL) techniques can be effective in reducing annotation costs for training deep models. However, their effectiveness in low- and high-budget scenarios seems to require different strategies, and achieving optimal results across varying budget scenarios remains a challenge. In this study, we introduce Dynamic Coverage & Margin mix (DCoM), a novel active learning approach designed to bridge this gap. Unlike existing strategies, DCoM dynamically adjusts its strategy, considering the competence of the current model. Through theoretical analysis and empirical evaluations on diverse datasets, including challenging computer vision tasks, we demonstrate DCoM's ability to overcome the cold start problem and consistently improve results across different budgetary constraints. Thus DCoM achieves state-of-the-art performance in both low- and high-budget regimes.", "date": "2024-07-01", "categories": []}
{"id": "2509.15805", "title": "Boosting Active Learning with Knowledge Transfer", "abstract": "Uncertainty estimation is at the core of Active Learning (AL). Most existing methods resort to complex auxiliary models and advanced training fashions to estimate uncertainty for unlabeled data. These models need special design and hence are difficult to train especially for domain tasks, such as Cryo-Electron Tomography (cryo-ET) classification in computational biology. To address this challenge, we propose a novel method using knowledge transfer to boost uncertainty estimation in AL. Specifically, we exploit the teacher-student mode where the teacher is the task model in AL and the student is an auxiliary model that learns from the teacher. We train the two models simultaneously in each AL cycle and adopt a certain distance between the model outputs to measure uncertainty for unlabeled data. The student model is task-agnostic and does not rely on special training fashions (e.g. adversarial), making our method suitable for various tasks. More importantly, we demonstrate that data uncertainty is not tied to concrete value of task loss but closely related to the upper-bound of task loss. We conduct extensive experiments to validate the proposed method on classical computer vision tasks and cryo-ET challenges. The results demonstrate its efficacy and efficiency.", "date": "2025-09-19", "categories": []}
{"id": "2509.08530", "title": "Data Skeleton Learning: Scalable Active Clustering with Sparse Graph Structures", "abstract": "In this work, we focus on the efficiency and scalability of pairwise constraint-based active clustering, crucial for processing large-scale data in applications such as data mining, knowledge annotation, and AI model pre-training. Our goals are threefold: (1) to reduce computational costs for iterative clustering updates; (2) to enhance the impact of user-provided constraints to minimize annotation requirements for precise clustering; and (3) to cut down memory usage in practical deployments. To achieve these aims, we propose a graph-based active clustering algorithm that utilizes two sparse graphs: one for representing relationships between data (our proposed data skeleton) and another for updating this data skeleton. These two graphs work in concert, enabling the refinement of connected subgraphs within the data skeleton to create nested clusters. Our empirical analysis confirms that the proposed algorithm consistently facilitates more accurate clustering with dramatically less input of user-provided constraints, and outperforms its counterparts in terms of computational performance and scalability, while maintaining robustness across various distance metrics.", "date": "2025-09-10", "categories": []}
{"id": "2301.08789", "title": "Active Learning of Piecewise Gaussian Process Surrogates", "abstract": "Active learning of Gaussian process (GP) surrogates has been useful for optimizing experimental designs for physical/computer simulation experiments, and for steering data acquisition schemes in machine learning. In this paper, we develop a method for active learning of piecewise, Jump GP surrogates. Jump GPs are continuous within, but discontinuous across, regions of a design space, as required for applications spanning autonomous materials design, configuration of smart factory systems, and many others. Although our active learning heuristics are appropriated from strategies originally designed for ordinary GPs, we demonstrate that additionally accounting for model bias, as opposed to the usual model uncertainty, is essential in the Jump GP context. Toward that end, we develop an estimator for bias and variance of Jump GP models. Illustrations, and evidence of the advantage of our proposed methods, are provided on a suite of synthetic benchmarks, and real-simulation experiments of varying complexity.", "date": "2023-01-20", "categories": []}
{"id": "2509.06978", "title": "A Kriging-HDMR-based surrogate model with sample pool-free active learning strategy for reliability analysis", "abstract": "In reliability engineering, conventional surrogate models encounter the \"curse of dimensionality\" as the number of random variables increases. While the active learning Kriging surrogate approaches with high-dimensional model representation (HDMR) enable effective approximation of high-dimensional functions and are widely applied to optimization problems, there are rare studies specifically focused on reliability analysis, which prioritizes prediction accuracy in critical regions over uniform accuracy across the entire domain. This study develops an active learning surrogate model method based on the Kriging-HDMR modeling for reliability analysis. The proposed approach facilitates the approximation of high-dimensional limit state functions through a composite representation constructed from multiple low-dimensional sub-surrogate models. The architecture of the surrogate modeling framework comprises three distinct stages: developing single-variable sub-surrogate models for all random variables, identifying the requirements for coupling-variable sub-surrogate models, and constructing the coupling-variable sub-surrogate models. Optimization mathematical models for selection of design of experiment samples are formulated based on each stage's characteristics, with objectives incorporating uncertainty variance, predicted mean, sample location and inter-sample distances. A candidate sample pool-free approach is adopted to achieve the selection of informative samples. Numerical experiments demonstrate that the proposed method achieves high computational efficiency while maintaining strong predictive accuracy in solving high-dimensional reliability problems.", "date": "2025-08-30", "categories": []}
{"id": "2509.07512", "title": "ALLabel: Three-stage Active Learning for LLM-based Entity Recognition using Demonstration Retrieval", "abstract": "Many contemporary data-driven research efforts in the natural sciences, such as chemistry and materials science, require large-scale, high-performance entity recognition from scientific datasets. Large language models (LLMs) have increasingly been adopted to solve the entity recognition task, with the same trend being observed on all-spectrum NLP tasks. The prevailing entity recognition LLMs rely on fine-tuned technology, yet the fine-tuning process often incurs significant cost. To achieve a best performance-cost trade-off, we propose ALLabel, a three-stage framework designed to select the most informative and representative samples in preparing the demonstrations for LLM modeling. The annotated examples are used to construct a ground-truth retrieval corpus for LLM in-context learning. By sequentially employing three distinct active learning strategies, ALLabel consistently outperforms all baselines under the same annotation budget across three specialized domain datasets. Experimental results also demonstrate that selectively annotating only 5\\%-10\\% of the dataset with ALLabel can achieve performance comparable to the method annotating the entire dataset. Further analyses and ablation studies verify the effectiveness and generalizability of our proposal.", "date": "2025-09-09", "categories": []}
{"id": "2509.02130", "title": "Online Identification of IT Systems through Active Causal Learning", "abstract": "Identifying a causal model of an IT system is fundamental to many branches of systems engineering and operation. Such a model can be used to predict the effects of control actions, optimize operations, diagnose failures, detect intrusions, etc., which is central to achieving the longstanding goal of automating network and system management tasks. Traditionally, causal models have been designed and maintained by domain experts. This, however, proves increasingly challenging with the growing complexity and dynamism of modern IT systems. In this paper, we present the first principled method for online, data-driven identification of an IT system in the form of a causal model. The method, which we call active causal learning, estimates causal functions that capture the dependencies among system variables in an iterative fashion using Gaussian process regression based on system measurements, which are collected through a rollout-based intervention policy. We prove that this method is optimal in the Bayesian sense and that it produces effective interventions. Experimental validation on a testbed shows that our method enables accurate identification of a causal system model while inducing low interference with system operations.", "date": "2025-09-02", "categories": []}
{"id": "2411.17941", "title": "Multi-Label Bayesian Active Learning with Inter-Label Relationships", "abstract": "The primary challenge of multi-label active learning, differing it from multi-class active learning, lies in assessing the informativeness of an indefinite number of labels while also accounting for the inherited label correlation. Existing studies either require substantial computational resources to leverage correlations or fail to fully explore label dependencies. Additionally, real-world scenarios often require addressing intrinsic biases stemming from imbalanced data distributions. In this paper, we propose a new multi-label active learning strategy to address both challenges. Our method incorporates progressively updated positive and negative correlation matrices to capture co-occurrence and disjoint relationships within the label space of annotated samples, enabling a holistic assessment of uncertainty rather than treating labels as isolated elements. Furthermore, alongside diversity, our model employs ensemble pseudo labeling and beta scoring rules to address data imbalances. Extensive experiments on four realistic datasets demonstrate that our strategy consistently achieves more reliable and superior performance, compared to several established methods.", "date": "2024-11-26", "categories": []}
{"id": "2509.03749", "title": "Mapping on a Budget: Optimizing Spatial Data Collection for ML", "abstract": "In applications across agriculture, ecology, and human development, machine learning with satellite imagery (SatML) is limited by the sparsity of labeled training data. While satellite data cover the globe, labeled training datasets for SatML are often small, spatially clustered, and collected for other purposes (e.g., administrative surveys or field measurements). Despite the pervasiveness of this issue in practice, past SatML research has largely focused on new model architectures and training algorithms to handle scarce training data, rather than modeling data conditions directly. This leaves scientists and policymakers who wish to use SatML for large-scale monitoring uncertain about whether and how to collect additional data to maximize performance. Here, we present the first problem formulation for the optimization of spatial training data in the presence of heterogeneous data collection costs and realistic budget constraints, as well as novel methods for addressing this problem. In experiments simulating different problem settings across three continents and four tasks, our strategies reveal substantial gains from sample optimization. Further experiments delineate settings for which optimized sampling is particularly effective. The problem formulation and methods we introduce are designed to generalize across application domains for SatML; we put special emphasis on a specific problem setting where our coauthors can immediately use our findings to augment clustered agricultural surveys for SatML monitoring in Togo.", "date": "2025-09-03", "categories": []}
{"id": "2509.04363", "title": "When three experiments are better than two: Avoiding intractable correlated aleatoric uncertainty by leveraging a novel bias--variance tradeoff", "abstract": "Real-world experimental scenarios are characterized by the presence of heteroskedastic aleatoric uncertainty, and this uncertainty can be correlated in batched settings. The bias--variance tradeoff can be used to write the expected mean squared error between a model distribution and a ground-truth random variable as the sum of an epistemic uncertainty term, the bias squared, and an aleatoric uncertainty term. We leverage this relationship to propose novel active learning strategies that directly reduce the bias between experimental rounds, considering model systems both with and without noise. Finally, we investigate methods to leverage historical data in a quadratic manner through the use of a novel cobias--covariance relationship, which naturally proposes a mechanism for batching through an eigendecomposition strategy. When our difference-based method leveraging the cobias--covariance relationship is utilized in a batched setting (with a quadratic estimator), we outperform a number of canonical methods including BALD and Least Confidence.", "date": "2025-09-04", "categories": []}
{"id": "2503.00479", "title": "Bayesian Active Learning for Multi-Criteria Comparative Judgement in Educational Assessment", "abstract": "Comparative Judgement (CJ) provides an alternative assessment approach by evaluating work holistically rather than breaking it into discrete criteria. This method leverages human ability to make nuanced comparisons, yielding more reliable and valid assessments. CJ aligns with real-world evaluations, where overall quality emerges from the interplay of various elements. However, rubrics remain widely used in education, offering structured criteria for grading and detailed feedback. This creates a gap between CJ's holistic ranking and the need for criterion-based performance breakdowns.\n  This paper addresses this gap using a Bayesian approach. We build on Bayesian CJ (BCJ) by Gray et al., which directly models preferences instead of using likelihoods over total scores, allowing for expected ranks with uncertainty estimation. Their entropy-based active learning method selects the most informative pairwise comparisons for assessors. We extend BCJ to handle multiple independent learning outcome (LO) components, defined by a rubric, enabling both holistic and component-wise predictive rankings with uncertainty estimates. Additionally, we propose a method to aggregate entropies and identify the most informative comparison for assessors. Experiments on synthetic and real data demonstrate our method's effectiveness. Finally, we address a key limitation of BCJ, which is the inability to quantify assessor agreement. We show how to derive agreement levels, enhancing transparency in assessment.", "date": "2025-03-01", "categories": []}
{"id": "2507.15976", "title": "Efficient dataset construction using active learning and uncertainty-aware neural networks for plasma turbulent transport surrogate models", "abstract": "This work demonstrates a proof-of-principle for using uncertainty-aware architectures, in combination with active learning techniques and an in-the-loop physics simulation code as a data labeller, to construct efficient datasets for data-driven surrogate model generation. Building off of a previous proof-of-principle successfully demonstrating training set reduction on static pre-labelled datasets, using the ADEPT framework, this strategy was applied again to the plasma turbulent transport problem within tokamak fusion plasmas, specifically the QuaLiKiz quasilinear electrostatic gyrokinetic turbulent transport code. While QuaLiKiz provides relatively fast evaluations, this study specifically targeted small datasets to serve as a proxy for more expensive codes, such as CGYRO or GENE. The newly implemented algorithm uses the SNGP architecture for the classification component of the problem and the BNN-NCP architecture for the regression component, training models for all turbulent modes (ITG, TEM, ETG) and all transport fluxes ($Q_e$, $Q_i$, $\\Gamma_e$, $\\Gamma_i$, and $\\Pi_i$) described by the general QuaLiKiz output. With 45 active learning iterations, moving from a small initial training set of $10^{2}$ to a final set of $10^{4}$, the resulting models reached a $F_1$ classification performance of ~0.8 and a $R^2$ regression performance of ~0.75 on an independent test set across all outputs. This extrapolates to reaching the same performance and efficiency as the previous ADEPT pipeline, although on a problem with 1 extra input dimension. While the improvement rate achieved in this implementation diminishes faster than expected, the overall technique is formulated with components that can be upgraded and generalized to many surrogate modeling applications beyond plasma turbulent transport predictions.", "date": "2025-07-21", "categories": []}
{"id": "2509.00704", "title": "Why Pool When You Can Flow? Active Learning with GFlowNets", "abstract": "The scalability of pool-based active learning is limited by the computational cost of evaluating large unlabeled datasets, a challenge that is particularly acute in virtual screening for drug discovery. While active learning strategies such as Bayesian Active Learning by Disagreement (BALD) prioritize informative samples, it remains computationally intensive when scaled to libraries containing billions samples. In this work, we introduce BALD-GFlowNet, a generative active learning framework that circumvents this issue. Our method leverages Generative Flow Networks (GFlowNets) to directly sample objects in proportion to the BALD reward. By replacing traditional pool-based acquisition with generative sampling, BALD-GFlowNet achieves scalability that is independent of the size of the unlabeled pool. In our virtual screening experiment, we show that BALD-GFlowNet achieves a performance comparable to that of standard BALD baseline while generating more structurally diverse molecules, offering a promising direction for efficient and scalable molecular discovery.", "date": "2025-08-31", "categories": []}
{"id": "2508.21263", "title": "Deep Active Learning for Lung Disease Severity Classification from Chest X-rays: Learning with Less Data in the Presence of Class Imbalance", "abstract": "To reduce the amount of required labeled data for lung disease severity classification from chest X-rays (CXRs) under class imbalance, this study applied deep active learning with a Bayesian Neural Network (BNN) approximation and weighted loss function. This retrospective study collected 2,319 CXRs from 963 patients (mean age, 59.2 $\\pm$ 16.6 years; 481 female) at Emory Healthcare affiliated hospitals between January and November 2020. All patients had clinically confirmed COVID-19. Each CXR was independently labeled by 3 to 6 board-certified radiologists as normal, moderate, or severe. A deep neural network with Monte Carlo Dropout was trained using active learning to classify disease severity. Various acquisition functions were used to iteratively select the most informative samples from an unlabeled pool. Performance was evaluated using accuracy, area under the receiver operating characteristic curve (AU ROC), and area under the precision-recall curve (AU PRC). Training time and acquisition time were recorded. Statistical analysis included descriptive metrics and performance comparisons across acquisition strategies. Entropy Sampling achieved 93.7% accuracy (AU ROC, 0.91) in binary classification (normal vs. diseased) using 15.4% of the training data. In the multi-class setting, Mean STD sampling achieved 70.3% accuracy (AU ROC, 0.86) using 23.1% of the labeled data. These methods outperformed more complex and computationally expensive acquisition functions and significantly reduced labeling needs. Deep active learning with BNN approximation and weighted loss effectively reduces labeled data requirements while addressing class imbalance, maintaining or exceeding diagnostic performance.", "date": "2025-08-28", "categories": []}
{"id": "2508.20528", "title": "Learning What is Worth Learning: Active and Sequential Domain Adaptation for Multi-modal Gross Tumor Volume Segmentation", "abstract": "Accurate gross tumor volume segmentation on multi-modal medical data is critical for radiotherapy planning in nasopharyngeal carcinoma and glioblastoma. Recent advances in deep neural networks have brought promising results in medical image segmentation, leading to an increasing demand for labeled data. Since labeling medical images is time-consuming and labor-intensive, active learning has emerged as a solution to reduce annotation costs by selecting the most informative samples to label and adapting high-performance models with as few labeled samples as possible. Previous active domain adaptation (ADA) methods seek to minimize sample redundancy by selecting samples that are farthest from the source domain. However, such one-off selection can easily cause negative transfer, and access to source medical data is often limited. Moreover, the query strategy for multi-modal medical data remains unexplored. In this work, we propose an active and sequential domain adaptation framework for dynamic multi-modal sample selection in ADA. We derive a query strategy to prioritize labeling and training on the most valuable samples based on their informativeness and representativeness. Empirical validation on diverse gross tumor volume segmentation tasks demonstrates that our method achieves favorable segmentation performance, significantly outperforming state-of-the-art ADA methods. Code is available at the git repository: \\href{https://github.com/Hiyoochan/mmActS}{mmActS}.", "date": "2025-08-28", "categories": []}
{"id": "2508.19906", "title": "Streamlining the Development of Active Learning Methods in Real-World Object Detection", "abstract": "Active learning (AL) for real-world object detection faces computational and reliability challenges that limit practical deployment. Developing new AL methods requires training multiple detectors across iterations to compare against existing approaches. This creates high costs for autonomous driving datasets where the training of one detector requires up to 282 GPU hours. Additionally, AL method rankings vary substantially across validation sets, compromising reliability in safety-critical transportation systems. We introduce object-based set similarity ($\\mathrm{OSS}$), a metric that addresses these challenges. $\\mathrm{OSS}$ (1) quantifies AL method effectiveness without requiring detector training by measuring similarity between training sets and target domains using object-level features. This enables the elimination of ineffective AL methods before training. Furthermore, $\\mathrm{OSS}$ (2) enables the selection of representative validation sets for robust evaluation. We validate our similarity-based approach on three autonomous driving datasets (KITTI, BDD100K, CODA) using uncertainty-based AL methods as a case study with two detector architectures (EfficientDet, YOLOv3). This work is the first to unify AL training and evaluation strategies in object detection based on object similarity. $\\mathrm{OSS}$ is detector-agnostic, requires only labeled object crops, and integrates with existing AL pipelines. This provides a practical framework for deploying AL in real-world applications where computational efficiency and evaluation reliability are critical. Code is available at https://mos-ks.github.io/publications/.", "date": "2025-08-27", "categories": []}
{"id": "2508.19019", "title": "Metric Matters: A Formal Evaluation of Similarity Measures in Active Learning for Cyber Threat Intelligence", "abstract": "Advanced Persistent Threats (APTs) pose a severe challenge to cyber defense due to their stealthy behavior and the extreme class imbalance inherent in detection datasets. To address these issues, we propose a novel active learning-based anomaly detection framework that leverages similarity search to iteratively refine the decision space. Built upon an Attention-Based Autoencoder, our approach uses feature-space similarity to identify normal-like and anomaly-like instances, thereby enhancing model robustness with minimal oracle supervision. Crucially, we perform a formal evaluation of various similarity measures to understand their influence on sample selection and anomaly ranking effectiveness. Through experiments on diverse datasets, including DARPA Transparent Computing APT traces, we demonstrate that the choice of similarity metric significantly impacts model convergence, anomaly detection accuracy, and label efficiency. Our results offer actionable insights for selecting similarity functions in active learning pipelines tailored for threat intelligence and cyber defense.", "date": "2025-08-26", "categories": []}
{"id": "2411.14618", "title": "Active Learning-Based Optimization of Hydroelectric Turbine Startup to Minimize Fatigue Damage", "abstract": "Hydro-generating units (HGUs) play a crucial role in integrating intermittent renewable energy sources into the power grid due to their flexible operational capabilities. This evolving role has led to an increase in transient events, such as startups, which impose significant stresses on turbines, leading to increased turbine fatigue and a reduced operational lifespan. Consequently, optimizing startup sequences to minimize stresses is vital for hydropower utilities. However, this task is challenging, as stress measurements on prototypes can be expensive and time-consuming. To tackle this challenge, we propose an innovative automated approach to optimize the startup parameters of HGUs with a limited budget of measured startup sequences. Our method combines active learning and black-box optimization techniques, utilizing virtual strain sensors and dynamic simulations of HGUs. This approach was tested in real-time during an on-site measurement campaign on an instrumented Francis turbine prototype. The results demonstrate that our algorithm successfully identified an optimal startup sequence using only seven measured sequences. It achieves a remarkable 42% reduction in the maximum strain cycle amplitude compared to the standard startup sequence. This study paves the way for more efficient HGU startup optimization, potentially extending their operational lifespans.", "date": "2024-11-21", "categories": []}
{"id": "2508.17849", "title": "Box-Level Class-Balanced Sampling for Active Object Detection", "abstract": "Training deep object detectors demands expensive bounding box annotation. Active learning (AL) is a promising technique to alleviate the annotation burden. Performing AL at box-level for object detection, i.e., selecting the most informative boxes to label and supplementing the sparsely-labelled image with pseudo labels, has been shown to be more cost-effective than selecting and labelling the entire image. In box-level AL for object detection, we observe that models at early stage can only perform well on majority classes, making the pseudo labels severely class-imbalanced. We propose a class-balanced sampling strategy to select more objects from minority classes for labelling, so as to make the final training data, \\ie, ground truth labels obtained by AL and pseudo labels, more class-balanced to train a better model. We also propose a task-aware soft pseudo labelling strategy to increase the accuracy of pseudo labels. We evaluate our method on public benchmarking datasets and show that our method achieves state-of-the-art performance.", "date": "2025-08-25", "categories": []}
{"id": "2210.06873", "title": "Data augmentation on-the-fly and active learning in data stream classification", "abstract": "There is an emerging need for predictive models to be trained on-the-fly, since in numerous machine learning applications data are arriving in an online fashion. A critical challenge encountered is that of limited availability of ground truth information (e.g., labels in classification tasks) as new data are observed one-by-one online, while another significant challenge is that of class imbalance. This work introduces the novel Augmented Queues method, which addresses the dual-problem by combining in a synergistic manner online active learning, data augmentation, and a multi-queue memory to maintain separate and balanced queues for each class. We perform an extensive experimental study using image and time-series augmentations, in which we examine the roles of the active learning budget, memory size, imbalance level, and neural network type. We demonstrate two major advantages of Augmented Queues. First, it does not reserve additional memory space as the generation of synthetic data occurs only at training times. Second, learning models have access to more labelled data without the need to increase the active learning budget and / or the original memory size. Learning on-the-fly poses major challenges which, typically, hinder the deployment of learning models. Augmented Queues significantly improves the performance in terms of learning quality and speed. Our code is made publicly available.", "date": "2022-10-13", "categories": []}
{"id": "2502.06209", "title": "Enhancing Cost Efficiency in Active Learning with Candidate Set Query", "abstract": "This paper introduces a cost-efficient active learning (AL) framework for classification, featuring a novel query design called candidate set query. Unlike traditional AL queries requiring the oracle to examine all possible classes, our method narrows down the set of candidate classes likely to include the ground-truth class, significantly reducing the search space and labeling cost. Moreover, we leverage conformal prediction to dynamically generate small yet reliable candidate sets, adapting to model enhancement over successive AL rounds. To this end, we introduce an acquisition function designed to prioritize data points that offer high information gain at lower cost. Empirical evaluations on CIFAR-10, CIFAR-100, and ImageNet64x64 demonstrate the effectiveness and scalability of our framework. Notably, it reduces labeling cost by 48% on ImageNet64x64. The project page can be found at https://yehogwon.github.io/csq-al.", "date": "2025-02-10", "categories": []}
{"id": "2507.04630", "title": "Learn 3D VQA Better with Active Selection and Reannotation", "abstract": "3D Visual Question Answering (3D VQA) is crucial for enabling models to perceive the physical world and perform spatial reasoning. In 3D VQA, the free-form nature of answers often leads to improper annotations that can confuse or mislead models when training on the entire dataset. While other text generation tasks can mitigate this issue by learning on large-scale datasets, the scarcity of 3D scene data enlarges the negative effect of misleading annotations. Although active learning strategies can select valuable instances for training, they fail to identify and resolve misleading labels, which the oracle inevitably provides in practice. To address this issue, we propose a multi-turn interactive active learning strategy. This strategy selects data based on models' semantic uncertainty to form a solid knowledge foundation more effectively and actively requests reannotation from an oracle to resolve potentially misleading labels. For uncertainty assessment, we utilize a variance-based metric that takes semantic relationships between terms into consideration, thus avoiding the uniform inter-class similarity assumption of previous assessment metrics. Extensive experiments exhibit better model performance and a substantial reduction in training costs, with a halving of training costs for achieving relatively high accuracy. The code is available at https://github.com/fz-zsl/AQuA.", "date": "2025-07-07", "categories": []}
{"id": "2308.08029", "title": "Sophisticated Learning: A novel algorithm for active learning during model-based planning", "abstract": "We introduce Sophisticated Learning (SL), a planning-to-learn algorithm that embeds active parameter learning inside the Sophisticated Inference (SI) tree-search framework of Active Inference. Unlike SI -- which optimizes beliefs about hidden states -- SL also updates beliefs about model parameters within each simulated branch, enabling counterfactual reasoning about how future observations would improve subsequent planning.\n  We compared SL with Bayes-adaptive Reinforcement Learning (BARL) agents as well as with its parent algorithm, SI. Using a biologically inspired seasonal foraging task in which resources shift probabilistically over a 10x10 grid, we designed experiments that forced agents to balance probabilistic reward harvesting against information gathering.\n  In early trials, where rapid learning is vital, SL agents survive, on average, 8.2% longer than SI and 35% longer than Bayes-adaptive Reinforcement Learning. While both SL and SI showed equal convergence performance, SL reached this convergence 40% faster than SI. Additionally, SL showed robust out-performance of other algorithms in altered environment configurations.\n  Our results show that incorporating active learning into multi-step planning materially improves decision making under radical uncertainty, and reinforces the broader utility of Active Inference for modeling biologically relevant behavior.", "date": "2023-08-15", "categories": []}
{"id": "2507.21846", "title": "Probabilistic Active Goal Recognition", "abstract": "In multi-agent environments, effective interaction hinges on understanding the beliefs and intentions of other agents. While prior work on goal recognition has largely treated the observer as a passive reasoner, Active Goal Recognition (AGR) focuses on strategically gathering information to reduce uncertainty. We adopt a probabilistic framework for Active Goal Recognition and propose an integrated solution that combines a joint belief update mechanism with a Monte Carlo Tree Search (MCTS) algorithm, allowing the observer to plan efficiently and infer the actor's hidden goal without requiring domain-specific knowledge. Through comprehensive empirical evaluation in a grid-based domain, we show that our joint belief update significantly outperforms passive goal recognition, and that our domain-independent MCTS performs comparably to our strong domain-specific greedy baseline. These results establish our solution as a practical and robust framework for goal inference, advancing the field toward more interactive and adaptive multi-agent systems.", "date": "2025-07-29", "categories": []}
{"id": "2508.08126", "title": "OFAL: An Oracle-Free Active Learning Framework", "abstract": "In the active learning paradigm, using an oracle to label data has always been a complex and expensive task, and with the emersion of large unlabeled data pools, it would be highly beneficial If we could achieve better results without relying on an oracle. This research introduces OFAL, an oracle-free active learning scheme that utilizes neural network uncertainty. OFAL uses the model's own uncertainty to transform highly confident unlabeled samples into informative uncertain samples. First, we start with separating and quantifying different parts of uncertainty and introduce Monte Carlo Dropouts as an approximation of the Bayesian Neural Network model. Secondly, by adding a variational autoencoder, we go on to generate new uncertain samples by stepping toward the uncertain part of latent space starting from a confidence seed sample. By generating these new informative samples, we can perform active learning and enhance the model's accuracy. Lastly, we try to compare and integrate our method with other widely used active learning sampling methods.", "date": "2025-08-11", "categories": []}
{"id": "2508.05681", "title": "Selection-Based Vulnerabilities: Clean-Label Backdoor Attacks in Active Learning", "abstract": "Active learning(AL), which serves as the representative label-efficient learning paradigm, has been widely applied in resource-constrained scenarios. The achievement of AL is attributed to acquisition functions, which are designed for identifying the most important data to label. Despite this success, one question remains unanswered: is AL safe? In this work, we introduce ALA, a practical and the first framework to utilize the acquisition function as the poisoning attack surface to reveal the weakness of active learning. Specifically, ALA optimizes imperceptibly poisoned inputs to exhibit high uncertainty scores, increasing their probability of being selected by acquisition functions. To evaluate ALA, we conduct extensive experiments across three datasets, three acquisition functions, and two types of clean-label backdoor triggers. Results show that our attack can achieve high success rates (up to 94%) even under low poisoning budgets (0.5%-1.0%) while preserving model utility and remaining undetectable to human annotators. Our findings remind active learning users: acquisition functions can be easily exploited, and active learning should be deployed with caution in trusted data scenarios.", "date": "2025-08-05", "categories": []}
{"id": "2508.04937", "title": "ALScope: A Unified Toolkit for Deep Active Learning", "abstract": "Deep Active Learning (DAL) reduces annotation costs by selecting the most informative unlabeled samples during training. As real-world applications become more complex, challenges stemming from distribution shifts (e.g., open-set recognition) and data imbalance have gained increasing attention, prompting the development of numerous DAL algorithms. However, the lack of a unified platform has hindered fair and systematic evaluation under diverse conditions. Therefore, we present a new DAL platform ALScope for classification tasks, integrating 10 datasets from computer vision (CV) and natural language processing (NLP), and 21 representative DAL algorithms, including both classical baselines and recent approaches designed to handle challenges such as distribution shifts and data imbalance. This platform supports flexible configuration of key experimental factors, ranging from algorithm and dataset choices to task-specific factors like out-of-distribution (OOD) sample ratio, and class imbalance ratio, enabling comprehensive and realistic evaluation. We conduct extensive experiments on this platform under various settings. Our findings show that: (1) DAL algorithms' performance varies significantly across domains and task settings; (2) in non-standard scenarios such as imbalanced and open-set settings, DAL algorithms show room for improvement and require further investigation; and (3) some algorithms achieve good performance, but require significantly longer selection time.", "date": "2025-08-06", "categories": []}
{"id": "2408.13491", "title": "ESA: Annotation-Efficient Active Learning for Semantic Segmentation", "abstract": "Active learning enhances annotation efficiency by selecting the most revealing samples for labeling, thereby reducing reliance on extensive human input. Previous methods in semantic segmentation have centered on individual pixels or small areas, neglecting the rich patterns in natural images and the power of advanced pre-trained models. To address these challenges, we propose three key contributions: Firstly, we introduce Entity-Superpixel Annotation (ESA), an innovative and efficient active learning strategy which utilizes a class-agnostic mask proposal network coupled with super-pixel grouping to capture local structural cues. Additionally, our method selects a subset of entities within each image of the target domain, prioritizing superpixels with high entropy to ensure comprehensive representation. Simultaneously, it focuses on a limited number of key entities, thereby optimizing for efficiency. By utilizing an annotator-friendly design that capitalizes on the inherent structure of images, our approach significantly outperforms existing pixel-based methods, achieving superior results with minimal queries, specifically reducing click cost by 98% and enhancing performance by 1.71%. For instance, our technique requires a mere 40 clicks for annotation, a stark contrast to the 5000 clicks demanded by conventional methods.", "date": "2024-08-24", "categories": []}
{"id": "2508.01957", "title": "Stochastic Encodings for Active Feature Acquisition", "abstract": "Active Feature Acquisition is an instance-wise, sequential decision making problem. The aim is to dynamically select which feature to measure based on current observations, independently for each test instance. Common approaches either use Reinforcement Learning, which experiences training difficulties, or greedily maximize the conditional mutual information of the label and unobserved features, which makes myopic acquisitions. To address these shortcomings, we introduce a latent variable model, trained in a supervised manner. Acquisitions are made by reasoning about the features across many possible unobserved realizations in a stochastic latent space. Extensive evaluation on a large range of synthetic and real datasets demonstrates that our approach reliably outperforms a diverse set of baselines.", "date": "2025-08-03", "categories": []}
{"id": "2508.03921", "title": "Active Learning and Transfer Learning for Anomaly Detection in Time-Series Data", "abstract": "This paper examines the effectiveness of combining active learning and transfer learning for anomaly detection in cross-domain time-series data. Our results indicate that there is an interaction between clustering and active learning and in general the best performance is achieved using a single cluster (in other words when clustering is not applied). Also, we find that adding new samples to the training set using active learning does improve model performance but that in general, the rate of improvement is slower than the results reported in the literature suggest. We attribute this difference to an improved experimental design where distinct data samples are used for the sampling and testing pools. Finally, we assess the ceiling performance of transfer learning in combination with active learning across several datasets and find that performance does initially improve but eventually begins to tail off as more target points are selected for inclusion in training. This tail-off in performance may indicate that the active learning process is doing a good job of sequencing data points for selection, pushing the less useful points towards the end of the selection process and that this tail-off occurs when these less useful points are eventually added. Taken together our results indicate that active learning is effective but that the improvement in model performance follows a linear flat function concerning the number of points selected and labelled.", "date": "2025-08-05", "categories": []}
{"id": "2508.00586", "title": "The Role of Active Learning in Modern Machine Learning", "abstract": "Even though Active Learning (AL) is widely studied, it is rarely applied in contexts outside its own scientific literature. We posit that the reason for this is AL's high computational cost coupled with the comparatively small lifts it is typically able to generate in scenarios with few labeled points. In this work we study the impact of different methods to combat this low data scenario, namely data augmentation (DA), semi-supervised learning (SSL) and AL. We find that AL is by far the least efficient method of solving the low data problem, generating a lift of only 1-4\\% over random sampling, while DA and SSL methods can generate up to 60\\% lift in combination with random sampling. However, when AL is combined with strong DA and SSL techniques, it surprisingly is still able to provide improvements. Based on these results, we frame AL not as a method to combat missing labels, but as the final building block to squeeze the last bits of performance out of data after appropriate DA and SSL methods as been applied.", "date": "2025-08-01", "categories": []}
{"id": "2507.23756", "title": "Improving annotator selection in Active Learning using a mood and fatigue-aware Recommender System", "abstract": "This study centers on overcoming the challenge of selecting the best annotators for each query in Active Learning (AL), with the objective of minimizing misclassifications. AL recognizes the challenges related to cost and time when acquiring labeled data, and decreases the number of labeled data needed. Nevertheless, there is still the necessity to reduce annotation errors, aiming to be as efficient as possible, to achieve the expected accuracy faster. Most strategies for query-annotator pairs do not consider internal factors that affect productivity, such as mood, attention, motivation, and fatigue levels. This work addresses this gap in the existing literature, by not only considering how the internal factors influence annotators (mood and fatigue levels) but also presenting a new query-annotator pair strategy, using a Knowledge-Based Recommendation System (RS). The RS ranks the available annotators, allowing to choose one or more to label the queried instance using their past accuracy values, and their mood and fatigue levels, as well as information about the instance queried. This work bases itself on existing literature on mood and fatigue influence on human performance, simulating annotators in a realistic manner, and predicting their performance with the RS. The results show that considering past accuracy values, as well as mood and fatigue levels reduces the number of annotation errors made by the annotators, and the uncertainty of the model through its training, when compared to not using internal factors. Accuracy and F1-score values were also better in the proposed approach, despite not being as substantial as the aforementioned. The methodologies and findings presented in this study begin to explore the open challenge of human cognitive factors affecting AL.", "date": "2025-07-31", "categories": []}
{"id": "2507.23771", "title": "Consensus-Driven Active Model Selection", "abstract": "The widespread availability of off-the-shelf machine learning models poses a challenge: which model, of the many available candidates, should be chosen for a given data analysis task? This question of model selection is traditionally answered by collecting and annotating a validation dataset -- a costly and time-intensive process. We propose a method for active model selection, using predictions from candidate models to prioritize the labeling of test data points that efficiently differentiate the best candidate. Our method, CODA, performs consensus-driven active model selection by modeling relationships between classifiers, categories, and data points within a probabilistic framework. The framework uses the consensus and disagreement between models in the candidate pool to guide the label acquisition process, and Bayesian inference to update beliefs about which model is best as more information is collected. We validate our approach by curating a collection of 26 benchmark tasks capturing a range of model selection scenarios. CODA outperforms existing methods for active model selection significantly, reducing the annotation effort required to discover the best model by upwards of 70% compared to the previous state-of-the-art. Code and data are available at https://github.com/justinkay/coda.", "date": "2025-07-31", "categories": []}
{"id": "2308.09301", "title": "Automata Learning from Preference and Equivalence Queries", "abstract": "Active automata learning from membership and equivalence queries is a foundational problem with numerous applications. We propose a novel variant of the active automata learning problem: actively learn finite automata using preference queries -- i.e., queries about the relative position of two sequences in a total order -- instead of membership queries. Our solution is REMAP, a novel algorithm which leverages a symbolic observation table along with unification and constraint solving to navigate a space of symbolic hypotheses (each representing a set of automata), and uses satisfiability-solving to construct a concrete automaton from a symbolic hypothesis. REMAP is guaranteed to correctly infer the minimal automaton with polynomial query complexity under exact equivalence queries, and achieves PAC-identification ($\\varepsilon$-approximate, with high probability) of the minimal automaton using sampling-based equivalence queries. Our empirical evaluations of REMAP on the task of learning reward machines for two reinforcement learning domains indicate REMAP scales to large automata and is effective at learning correct automata from consistent teachers, under both exact and sampling-based equivalence queries.", "date": "2023-08-18", "categories": []}
{"id": "2505.20930", "title": "MLMC-based Resource Adequacy Assessment with Active Learning Trained Surrogate Models", "abstract": "Multilevel Monte Carlo (MLMC) is a flexible and effective variance reduction technique for accelerating reliability assessments of complex power system. Recently, data-driven surrogate models have been proposed as lower-level models in the MLMC framework due to their high correlation and negligible execution time once trained. However, in resource adequacy assessments, pre-labeled datasets are typically unavailable. For large-scale systems, the efficiency gains from surrogate models are often offset by the substantial time required for labeling training data. Therefore, this paper introduces a speed metric that accounts for training time in evaluating MLMC efficiency. Considering the total time budget is limited, a vote-by-committee active learning approach is proposed to reduce the required labeling calls. A case study demonstrates that, within a given computational budget, active learning in combination with MLMC can result in a substantial reduction variance.", "date": "2025-05-27", "categories": []}
{"id": "2507.21521", "title": "Optimizing Active Learning in Vision-Language Models via Parameter-Efficient Uncertainty Calibration", "abstract": "Active Learning (AL) has emerged as a powerful approach for minimizing labeling costs by selectively sampling the most informative data for neural network model development. Effective AL for large-scale vision-language models necessitates addressing challenges in uncertainty estimation and efficient sampling given the vast number of parameters involved. In this work, we introduce a novel parameter-efficient learning methodology that incorporates uncertainty calibration loss within the AL framework. We propose a differentiable loss function that promotes uncertainty calibration for effectively selecting fewer and most informative data samples for fine-tuning. Through extensive experiments across several datasets and vision backbones, we demonstrate that our solution can match and exceed the performance of complex feature-based sampling techniques while being computationally very efficient. Additionally, we investigate the efficacy of Prompt learning versus Low-rank adaptation (LoRA) in sample selection, providing a detailed comparative analysis of these methods in the context of efficient AL.", "date": "2025-07-29", "categories": []}
{"id": "2507.20490", "title": "HIAL: A New Paradigm for Hypergraph Active Learning via Influence Maximization", "abstract": "In recent years, Hypergraph Neural Networks (HNNs) have demonstrated immense potential in handling complex systems with high-order interactions. However, acquiring large-scale, high-quality labeled data for these models is costly, making Active Learning (AL) a critical technique. Existing Graph Active Learning (GAL) methods, when applied to hypergraphs, often rely on techniques like \"clique expansion,\" which destroys the high-order structural information crucial to a hypergraph's success, thereby leading to suboptimal performance. To address this challenge, we introduce HIAL (Hypergraph Active Learning), a native active learning framework designed specifically for hypergraphs. We innovatively reformulate the Hypergraph Active Learning (HAL) problem as an Influence Maximization task. The core of HIAL is a dual-perspective influence function that, based on our novel \"High-Order Interaction-Aware (HOI-Aware)\" propagation mechanism, synergistically evaluates a node's feature-space coverage (via Magnitude of Influence, MoI) and its topological influence (via Expected Diffusion Value, EDV). We prove that this objective function is monotone and submodular, thus enabling the use of an efficient greedy algorithm with a formal (1-1/e) approximation guarantee. Extensive experiments on seven public datasets demonstrate that HIAL significantly outperforms state-of-the-art baselines in terms of performance, efficiency, generality, and robustness, establishing an efficient and powerful new paradigm for active learning on hypergraphs.", "date": "2025-07-28", "categories": []}
{"id": "2504.20011", "title": "Curiosity Driven Exploration to Optimize Structure-Property Learning in Microscopy", "abstract": "Rapidly determining structure-property correlations in materials is an important challenge in better understanding fundamental mechanisms and greatly assists in materials design. In microscopy, imaging data provides a direct measurement of the local structure, while spectroscopic measurements provide relevant functional property information. Deep kernel active learning approaches have been utilized to rapidly map local structure to functional properties in microscopy experiments, but are computationally expensive for multi-dimensional and correlated output spaces. Here, we present an alternative lightweight curiosity algorithm which actively samples regions with unexplored structure-property relations, utilizing a deep-learning based surrogate model for error prediction. We show that the algorithm outperforms random sampling for predicting properties from structures, and provides a convenient tool for efficient mapping of structure-property relationships in materials science.", "date": "2025-04-28", "categories": []}
{"id": "2507.19316", "title": "Human-AI Synergy in Adaptive Active Learning for Continuous Lithium Carbonate Crystallization Optimization", "abstract": "As demand for high-purity lithium surges with the growth of the electric vehicle (EV) industry, cost-effective extraction from lower-grade North American sources like the Smackover Formation is critical. These resources, unlike high-purity South American brines, require innovative purification techniques to be economically viable. Continuous crystallization is a promising method for producing battery-grade lithium carbonate, but its optimization is challenged by a complex parameter space and limited data. This study introduces a Human-in-the-Loop (HITL) assisted active learning framework to optimize the continuous crystallization of lithium carbonate. By integrating human expertise with data-driven insights, our approach accelerates the optimization of lithium extraction from challenging sources. Our results demonstrate the framework's ability to rapidly adapt to new data, significantly improving the process's tolerance to critical impurities like magnesium from the industry standard of a few hundred ppm to as high as 6000 ppm. This breakthrough makes the exploitation of low-grade, impurity-rich lithium resources feasible, potentially reducing the need for extensive pre-refinement processes. By leveraging artificial intelligence, we have refined operational parameters and demonstrated that lower-grade materials can be used without sacrificing product quality. This advancement is a significant step towards economically harnessing North America's vast lithium reserves, such as those in the Smackover Formation, and enhancing the sustainability of the global lithium supply chain.", "date": "2025-07-25", "categories": []}
{"id": "2507.18123", "title": "Actively evaluating and learning the distinctions that matter: Vaccine safety signal detection from emergency triage notes", "abstract": "The rapid development of COVID-19 vaccines has showcased the global communitys ability to combat infectious diseases. However, the need for post-licensure surveillance systems has grown due to the limited window for safety data collection in clinical trials and early widespread implementation. This study aims to employ Natural Language Processing techniques and Active Learning to rapidly develop a classifier that detects potential vaccine safety issues from emergency department notes. ED triage notes, containing expert, succinct vital patient information at the point of entry to health systems, can significantly contribute to timely vaccine safety signal surveillance. While keyword-based classification can be effective, it may yield false positives and demand extensive keyword modifications. This is exacerbated by the infrequency of vaccination-related ED presentations and their similarity to other reasons for ED visits. NLP offers a more accurate and efficient alternative, albeit requiring annotated data, which is often scarce in the medical field. Active learning optimizes the annotation process and the quality of annotated data, which can result in faster model implementation and improved model performance. This work combines active learning, data augmentation, and active learning and evaluation techniques to create a classifier that is used to enhance vaccine safety surveillance from ED triage notes.", "date": "2025-07-24", "categories": []}
{"id": "2310.11985", "title": "A Finite-Horizon Approach to Active Level Set Estimation", "abstract": "We consider the problem of active learning in the context of spatial sampling for level set estimation (LSE), where the goal is to localize all regions where a function of interest lies above/below a given threshold as quickly as possible. We present a finite-horizon search procedure to perform LSE in one dimension while optimally balancing both the final estimation error and the distance traveled for a fixed number of samples. A tuning parameter is used to trade off between the estimation accuracy and distance traveled. We show that the resulting optimization problem can be solved in closed form and that the resulting policy generalizes existing approaches to this problem. We then show how this approach can be used to perform level set estimation in higher dimensions under the popular Gaussian process model. Empirical results on synthetic data indicate that as the cost of travel increases, our method's ability to treat distance nonmyopically allows it to significantly improve on the state of the art. On real air quality data, our approach achieves roughly one fifth the estimation error at less than half the cost of competing algorithms.", "date": "2023-10-18", "categories": []}
{"id": "2411.00107", "title": "First, Learn What You Don't Know: Active Information Gathering for Driving at the Limits of Handling", "abstract": "Combining data-driven models that adapt online and model predictive control (MPC) has enabled effective control of nonlinear systems. However, when deployed on unstable systems, online adaptation may not be fast enough to ensure reliable simultaneous learning and control. For example, a controller on a vehicle executing highly dynamic maneuvers--such as drifting to avoid an obstacle--may push the vehicle's tires to their friction limits, destabilizing the vehicle and allowing modeling errors to quickly compound and cause a loss of control. To address this challenge, we present an active information gathering framework for identifying vehicle dynamics as quickly as possible. We propose an expressive vehicle dynamics model that leverages Bayesian last-layer meta-learning to enable rapid online adaptation. The model's uncertainty estimates are used to guide informative data collection and quickly improve the model prior to deployment. Dynamic drifting experiments on a Toyota Supra show that (i) the framework enables reliable control of a vehicle at the edge of stability, (ii) online adaptation alone may not suffice for zero-shot control and can lead to undesirable transient errors or spin-outs, and (iii) active data collection helps achieve reliable performance.", "date": "2024-10-31", "categories": []}
{"id": "2507.17359", "title": "Exploring Active Learning for Semiconductor Defect Segmentation", "abstract": "The development of X-Ray microscopy (XRM) technology has enabled non-destructive inspection of semiconductor structures for defect identification. Deep learning is widely used as the state-of-the-art approach to perform visual analysis tasks. However, deep learning based models require large amount of annotated data to train. This can be time-consuming and expensive to obtain especially for dense prediction tasks like semantic segmentation. In this work, we explore active learning (AL) as a potential solution to alleviate the annotation burden. We identify two unique challenges when applying AL on semiconductor XRM scans: large domain shift and severe class-imbalance. To address these challenges, we propose to perform contrastive pretraining on the unlabelled data to obtain the initialization weights for each AL cycle, and a rareness-aware acquisition function that favors the selection of samples containing rare classes. We evaluate our method on a semiconductor dataset that is compiled from XRM scans of high bandwidth memory structures composed of logic and memory dies, and demonstrate that our method achieves state-of-the-art performance.", "date": "2025-07-23", "categories": []}
{"id": "2507.17367", "title": "Exploring Spatial Diversity for Region-based Active Learning", "abstract": "State-of-the-art methods for semantic segmentation are based on deep neural networks trained on large-scale labeled datasets. Acquiring such datasets would incur large annotation costs, especially for dense pixel-level prediction tasks like semantic segmentation. We consider region-based active learning as a strategy to reduce annotation costs while maintaining high performance. In this setting, batches of informative image regions instead of entire images are selected for labeling. Importantly, we propose that enforcing local spatial diversity is beneficial for active learning in this case, and to incorporate spatial diversity along with the traditional active selection criterion, e.g., data sample uncertainty, in a unified optimization framework for region-based active learning. We apply this framework to the Cityscapes and PASCAL VOC datasets and demonstrate that the inclusion of spatial diversity effectively improves the performance of uncertainty-based and feature diversity-based active learning methods. Our framework achieves $95\\%$ performance of fully supervised methods with only $5-9\\%$ of the labeled pixels, outperforming all state-of-the-art region-based active learning methods for semantic segmentation.", "date": "2025-07-23", "categories": []}
{"id": "2310.14968", "title": "Bayesian Active Learning in the Presence of Nuisance Parameters", "abstract": "In many settings, such as scientific inference, optimization, and transfer learning, the learner has a well-defined objective, which can be treated as estimation of a target parameter, and no intrinsic interest in characterizing the entire data-generating process. Usually, the learner must also contend with additional sources of uncertainty or variables -- with nuisance parameters. Bayesian active learning, or sequential optimal experimental design, can straightforwardly accommodate the presence of nuisance parameters, and so is a natural active learning framework for such problems. However, the introduction of nuisance parameters can lead to bias in the Bayesian learner's estimate of the target parameters, a phenomenon we refer to as negative interference. We characterize the threat of negative interference and how it fundamentally changes the nature of the Bayesian active learner's task. We show that the extent of negative interference can be extremely large, and that accurate estimation of the nuisance parameters is critical to reducing it. The Bayesian active learner is confronted with a dilemma: whether to spend a finite acquisition budget in pursuit of estimation of the target or of the nuisance parameters. Our setting encompasses Bayesian transfer learning as a special case, and our results shed light on the phenomenon of negative transfer between learning environments.", "date": "2023-10-23", "categories": []}
{"id": "2507.16424", "title": "PromptAL: Sample-Aware Dynamic Soft Prompts for Few-Shot Active Learning", "abstract": "Active learning (AL) aims to optimize model training and reduce annotation costs by selecting the most informative samples for labeling. Typically, AL methods rely on the empirical distribution of labeled data to define the decision boundary and perform uncertainty or diversity estimation, subsequently identifying potential high-quality samples. In few-shot scenarios, the empirical distribution often diverges significantly from the target distribution, causing the decision boundary to shift away from its optimal position. However, existing methods overlook the role of unlabeled samples in enhancing the empirical distribution to better align with the target distribution, resulting in a suboptimal decision boundary and the selection of samples that inadequately represent the target distribution. To address this, we propose a hybrid AL framework, termed \\textbf{PromptAL} (Sample-Aware Dynamic Soft \\textbf{Prompts} for Few-Shot \\textbf{A}ctive \\textbf{L}earning). This framework accounts for the contribution of each unlabeled data point in aligning the current empirical distribution with the target distribution, thereby optimizing the decision boundary. Specifically, PromptAL first leverages unlabeled data to construct sample-aware dynamic soft prompts that adjust the model's predictive distribution and decision boundary. Subsequently, based on the adjusted decision boundary, it integrates uncertainty estimation with both global and local diversity to select high-quality samples that more accurately represent the target distribution. Experimental results on six in-domain and three out-of-domain datasets show that PromptAL achieves superior performance over nine baselines. Our codebase is openly accessible.", "date": "2025-07-22", "categories": []}
{"id": "2507.14217", "title": "Geometry-Aware Active Learning of Pattern Rankings via Choquet-Based Aggregation", "abstract": "We address the pattern explosion problem in pattern mining by proposing an interactive learning framework that combines nonlinear utility aggregation with geometry-aware query selection. Our method models user preferences through a Choquet integral over multiple interestingness measures and exploits the geometric structure of the version space to guide the selection of informative comparisons. A branch-and-bound strategy with tight distance bounds enables efficient identification of queries near the decision boundary. Experiments on UCI datasets show that our approach outperforms existing methods such as ChoquetRank, achieving better ranking accuracy with fewer user interactions.", "date": "2025-07-16", "categories": []}
{"id": "2507.15381", "title": "To Label or Not to Label: PALM -- A Predictive Model for Evaluating Sample Efficiency in Active Learning Models", "abstract": "Active learning (AL) seeks to reduce annotation costs by selecting the most informative samples for labeling, making it particularly valuable in resource-constrained settings. However, traditional evaluation methods, which focus solely on final accuracy, fail to capture the full dynamics of the learning process. To address this gap, we propose PALM (Performance Analysis of Active Learning Models), a unified and interpretable mathematical model that characterizes AL trajectories through four key parameters: achievable accuracy, coverage efficiency, early-stage performance, and scalability. PALM provides a predictive description of AL behavior from partial observations, enabling the estimation of future performance and facilitating principled comparisons across different strategies. We validate PALM through extensive experiments on CIFAR-10/100 and ImageNet-50/100/200, covering a wide range of AL methods and self-supervised embeddings. Our results demonstrate that PALM generalizes effectively across datasets, budgets, and strategies, accurately predicting full learning curves from limited labeled data. Importantly, PALM reveals crucial insights into learning efficiency, data space coverage, and the scalability of AL methods. By enabling the selection of cost-effective strategies and predicting performance under tight budget constraints, PALM lays the basis for more systematic, reproducible, and data-efficient evaluation of AL in both research and real-world applications. The code is available at: https://github.com/juliamachnio/PALM.", "date": "2025-07-21", "categories": []}
{"id": "2504.12016", "title": "Active Human Feedback Collection via Neural Contextual Dueling Bandits", "abstract": "Collecting human preference feedback is often expensive, leading recent works to develop principled algorithms to select them more efficiently. However, these works assume that the underlying reward function is linear, an assumption that does not hold in many real-life applications, such as online recommendation and LLM alignment. To address this limitation, we propose Neural-ADB, an algorithm based on the neural contextual dueling bandit framework that provides a principled and practical method for collecting human preference feedback when the underlying latent reward function is non-linear. We theoretically show that when preference feedback follows the Bradley-Terry-Luce model, the worst sub-optimality gap of the policy learned by Neural-ADB decreases at a sub-linear rate as the preference dataset increases. Our experimental results on preference datasets further corroborate the effectiveness of Neural-ADB.", "date": "2025-04-16", "categories": []}
{"id": "2507.06537", "title": "A model-agnostic active learning approach for animal detection from camera traps", "abstract": "Smart data selection is becoming increasingly important in data-driven machine learning. Active learning offers a promising solution by allowing machine learning models to be effectively trained with optimal data including the most informative samples from large datasets. Wildlife data captured by camera traps are excessive in volume, requiring tremendous effort in data labelling and animal detection models training. Therefore, applying active learning to optimise the amount of labelled data would be a great aid in enabling automated wildlife monitoring and conservation. However, existing active learning techniques require that a machine learning model (i.e., an object detector) be fully accessible, limiting the applicability of the techniques. In this paper, we propose a model-agnostic active learning approach for detection of animals captured by camera traps. Our approach integrates uncertainty and diversity quantities of samples at both the object-based and image-based levels into the active learning sample selection process. We validate our approach in a benchmark animal dataset. Experimental results demonstrate that, using only 30% of the training data selected by our approach, a state-of-the-art animal detector can achieve a performance of equal or greater than that with the use of the complete training dataset.", "date": "2025-07-09", "categories": []}
{"id": "2502.16870", "title": "Distributionally Robust Active Learning for Gaussian Process Regression", "abstract": "Gaussian process regression (GPR) or kernel ridge regression is a widely used and powerful tool for nonlinear prediction. Therefore, active learning (AL) for GPR, which actively collects data labels to achieve an accurate prediction with fewer data labels, is an important problem. However, existing AL methods do not theoretically guarantee prediction accuracy for target distribution. Furthermore, as discussed in the distributionally robust learning literature, specifying the target distribution is often difficult. Thus, this paper proposes two AL methods that effectively reduce the worst-case expected error for GPR, which is the worst-case expectation in target distribution candidates. We show an upper bound of the worst-case expected squared error, which suggests that the error will be arbitrarily small by a finite number of data labels under mild conditions. Finally, we demonstrate the effectiveness of the proposed methods through synthetic and real-world datasets.", "date": "2025-02-24", "categories": []}
{"id": "2507.02959", "title": "A Novel Active Learning Approach to Label One Million Unknown Malware Variants", "abstract": "Active learning for classification seeks to reduce the cost of labeling samples by finding unlabeled examples about which the current model is least certain and sending them to an annotator/expert to label. Bayesian theory can provide a probabilistic view of deep neural network models by asserting a prior distribution over model parameters and estimating the uncertainties by posterior distribution over these parameters. This paper proposes two novel active learning approaches to label one million malware examples belonging to different unknown modern malware families. The first model is Inception-V4+PCA combined with several support vector machine (SVM) algorithms (UTSVM, PSVM, SVM-GSU, TBSVM). The second model is Vision Transformer based Bayesian Neural Networks ViT-BNN. Our proposed ViT-BNN is a state-of-the-art active learning approach that differs from current methods and can apply to any particular task. The experiments demonstrate that the ViT-BNN is more stable and robust in handling uncertainty.", "date": "2025-06-30", "categories": []}
{"id": "2507.04929", "title": "ConBatch-BAL: Batch Bayesian Active Learning under Budget Constraints", "abstract": "Varying annotation costs among data points and budget constraints can hinder the adoption of active learning strategies in real-world applications. This work introduces two Bayesian active learning strategies for batch acquisition under constraints (ConBatch-BAL), one based on dynamic thresholding and one following greedy acquisition. Both select samples using uncertainty metrics computed via Bayesian neural networks. The dynamic thresholding strategy redistributes the budget across the batch, while the greedy one selects the top-ranked sample at each step, limited by the remaining budget. Focusing on scenarios with costly data annotation and geospatial constraints, we also release two new real-world datasets containing geolocated aerial images of buildings, annotated with energy efficiency or typology classes. The ConBatch-BAL strategies are benchmarked against a random acquisition baseline on these datasets under various budget and cost scenarios. The results show that the developed ConBatch-BAL strategies can reduce active learning iterations and data acquisition costs in real-world settings, and even outperform the unconstrained baseline solutions.", "date": "2025-07-07", "categories": []}
{"id": "2507.02109", "title": "Parametric Neural Amp Modeling with Active Learning", "abstract": "We introduce PANAMA, an active learning framework for the training of end-to-end parametric guitar amp models using a WaveNet-like architecture. With \\model, one can create a virtual amp by recording samples that are determined by an active learning strategy to use a minimum amount of datapoints (i.e., amp knob settings). We show that gradient-based optimization algorithms can be used to determine the optimal datapoints to sample, and that the approach helps under a constrained number of samples.", "date": "2025-07-02", "categories": []}
{"id": "2507.02593", "title": "Revisiting Active Learning under (Human) Label Variation", "abstract": "Access to high-quality labeled data remains a limiting factor in applied supervised learning. While label variation (LV), i.e., differing labels for the same instance, is common, especially in natural language processing, annotation frameworks often still rest on the assumption of a single ground truth. This overlooks human label variation (HLV), the occurrence of plausible differences in annotations, as an informative signal. Similarly, active learning (AL), a popular approach to optimizing the use of limited annotation budgets in training ML models, often relies on at least one of several simplifying assumptions, which rarely hold in practice when acknowledging HLV. In this paper, we examine foundational assumptions about truth and label nature, highlighting the need to decompose observed LV into signal (e.g., HLV) and noise (e.g., annotation error). We survey how the AL and (H)LV communities have addressed -- or neglected -- these distinctions and propose a conceptual framework for incorporating HLV throughout the AL loop, including instance selection, annotator choice, and label representation. We further discuss the integration of large language models (LLM) as annotators. Our work aims to lay a conceptual foundation for HLV-aware active learning, better reflecting the complexities of real-world annotation.", "date": "2025-07-03", "categories": []}
{"id": "2507.00474", "title": "ADAptation: Reconstruction-based Unsupervised Active Learning for Breast Ultrasound Diagnosis", "abstract": "Deep learning-based diagnostic models often suffer performance drops due to distribution shifts between training (source) and test (target) domains. Collecting and labeling sufficient target domain data for model retraining represents an optimal solution, yet is limited by time and scarce resources. Active learning (AL) offers an efficient approach to reduce annotation costs while maintaining performance, but struggles to handle the challenge posed by distribution variations across different datasets. In this study, we propose a novel unsupervised Active learning framework for Domain Adaptation, named ADAptation, which efficiently selects informative samples from multi-domain data pools under limited annotation budget. As a fundamental step, our method first utilizes the distribution homogenization capabilities of diffusion models to bridge cross-dataset gaps by translating target images into source-domain style. We then introduce two key innovations: (a) a hypersphere-constrained contrastive learning network for compact feature clustering, and (b) a dual-scoring mechanism that quantifies and balances sample uncertainty and representativeness. Extensive experiments on four breast ultrasound datasets (three public and one in-house/multi-center) across five common deep classifiers demonstrate that our method surpasses existing strong AL-based competitors, validating its effectiveness and generalization for clinical domain adaptation. The code is available at the anonymized link: https://github.com/miccai25-966/ADAptation.", "date": "2025-07-01", "categories": []}
{"id": "2506.15318", "title": "OpenPath: Open-Set Active Learning for Pathology Image Classification via Pre-trained Vision-Language Models", "abstract": "Pathology image classification plays a crucial role in accurate medical diagnosis and treatment planning. Training high-performance models for this task typically requires large-scale annotated datasets, which are both expensive and time-consuming to acquire. Active Learning (AL) offers a solution by iteratively selecting the most informative samples for annotation, thereby reducing the labeling effort. However, most AL methods are designed under the assumption of a closed-set scenario, where all the unannotated images belong to target classes. In real-world clinical environments, the unlabeled pool often contains a substantial amount of Out-Of-Distribution (OOD) data, leading to low efficiency of annotation in traditional AL methods. Furthermore, most existing AL methods start with random selection in the first query round, leading to a significant waste of labeling costs in open-set scenarios. To address these challenges, we propose OpenPath, a novel open-set active learning approach for pathological image classification leveraging a pre-trained Vision-Language Model (VLM). In the first query, we propose task-specific prompts that combine target and relevant non-target class prompts to effectively select In-Distribution (ID) and informative samples from the unlabeled pool. In subsequent queries, Diverse Informative ID Sampling (DIS) that includes Prototype-based ID candidate Selection (PIS) and Entropy-Guided Stochastic Sampling (EGSS) is proposed to ensure both purity and informativeness in a query, avoiding the selection of OOD samples. Experiments on two public pathology image datasets show that OpenPath significantly enhances the model's performance due to its high purity of selected samples, and outperforms several state-of-the-art open-set AL methods. The code is available at \\href{https://github.com/HiLab-git/OpenPath}{https://github.com/HiLab-git/OpenPath}..", "date": "2025-06-18", "categories": []}
{"id": "2506.22444", "title": "Active Learning for Forecasting Severity among Patients with Post Acute Sequelae of SARS-CoV-2", "abstract": "The long-term effects of Postacute Sequelae of SARS-CoV-2, known as PASC, pose a significant challenge to healthcare systems worldwide. Accurate identification of progression events, such as hospitalization and reinfection, is essential for effective patient management and resource allocation. However, traditional models trained on structured data struggle to capture the nuanced progression of PASC. In this study, we introduce the first publicly available cohort of 18 PASC patients, with text time series features based on Large Language Model Llama-3.1-70B-Instruct and clinical risk annotated by clinical expert. We propose an Active Attention Network to predict the clinical risk and identify progression events related to the risk. By integrating human expertise with active learning, we aim to enhance clinical risk prediction accuracy and enable progression events identification with fewer number of annotation. The ultimate goal is to improves patient care and decision-making for SARS-CoV-2 patient.", "date": "2025-06-11", "categories": []}
{"id": "2506.22645", "title": "Cost-effective Reduced-Order Modeling via Bayesian Active Learning", "abstract": "Machine Learning surrogates have been developed to accelerate solving systems dynamics of complex processes in different science and engineering applications. To faithfully capture governing systems dynamics, these methods rely on large training datasets, hence restricting their applicability in real-world problems. In this work, we propose BayPOD-AL, an active learning framework based on an uncertainty-aware Bayesian proper orthogonal decomposition (POD) approach, which aims to effectively learn reduced-order models from high-fidelity full-order models representing complex systems. Experimental results on predicting the temperature evolution over a rod demonstrate BayPOD-AL's effectiveness in suggesting the informative data and reducing computational cost related to constructing a training dataset compared to other uncertainty-guided active learning strategies. Furthermore, we demonstrate BayPOD-AL's generalizability and efficiency by evaluating its performance on a dataset of higher temporal resolution than the training dataset.", "date": "2025-06-27", "categories": []}
{"id": "2506.23342", "title": "ATGen: A Framework for Active Text Generation", "abstract": "Active learning (AL) has demonstrated remarkable potential in reducing the annotation effort required for training machine learning models. However, despite the surging popularity of natural language generation (NLG) tasks in recent years, the application of AL to NLG has been limited. In this paper, we introduce Active Text Generation (ATGen) - a comprehensive framework that bridges AL with text generation tasks, enabling the application of state-of-the-art AL strategies to NLG. Our framework simplifies AL-empowered annotation in NLG tasks using both human annotators and automatic annotation agents based on large language models (LLMs). The framework supports LLMs deployed as services, such as ChatGPT and Claude, or operated on-premises. Furthermore, ATGen provides a unified platform for smooth implementation and benchmarking of novel AL strategies tailored to NLG tasks. Finally, we present evaluation results for state-of-the-art AL strategies across diverse settings and multiple text generation tasks. We show that ATGen reduces both the effort of human annotators and costs associated with API calls to LLM-based annotation agents. The code of the framework is available on GitHub under the MIT license. The video presentation is available at http://atgen-video.nlpresearch.group", "date": "2025-06-29", "categories": []}
{"id": "2505.00460", "title": "Subspace-Distance-Enabled Active Learning for Efficient Data-Driven Model Reduction of Parametric Dynamical Systems", "abstract": "In situations where the solution of a high-fidelity dynamical system needs to be evaluated repeatedly, over a vast pool of parametric configurations and in absence of access to the underlying governing equations, data-driven model reduction techniques are preferable. We propose a novel active learning approach to build a parametric data-driven reduced-order model (ROM) by greedily picking the most important parameter samples from the parameter domain. As a result, during the ROM construction phase, the number of high-fidelity solutions dynamically grow in a principled fashion. The high-fidelity solution snapshots are expressed in several parameter-specific linear subspaces, with the help of proper orthogonal decomposition (POD), and the relative distance between these subspaces is used as a guiding mechanism to perform active learning. For successfully achieving this, we provide a distance measure to evaluate the similarity between pairs of linear subspaces with different dimensions, and also show that this distance measure is a metric. The usability of the proposed subspace-distance-enabled active learning (SDE-AL) framework is demonstrated by augmenting two existing non-intrusive reduced-order modeling approaches, and providing their active-learning-driven (ActLearn) extensions, namely, SDE-ActLearn-POD-KSNN, and SDE-ActLearn-POD-NN. Furthermore, we report positive results for two parametric physical models, highlighting the efficiency of the proposed SDE-AL approach.", "date": "2025-05-01", "categories": []}
{"id": "2410.02145", "title": "Active Learning of Deep Neural Networks via Gradient-Free Cutting Planes", "abstract": "Active learning methods aim to improve sample complexity in machine learning. In this work, we investigate an active learning scheme via a novel gradient-free cutting-plane training method for ReLU networks of arbitrary depth and develop a convergence theory. We demonstrate, for the first time, that cutting-plane algorithms, traditionally used in linear models, can be extended to deep neural networks despite their nonconvexity and nonlinear decision boundaries. Moreover, this training method induces the first deep active learning scheme known to achieve convergence guarantees, revealing a geometric contraction rate of the feasible set. We exemplify the effectiveness of our proposed active learning method against popular deep active learning baselines via both synthetic data experiments and sentimental classification task on real datasets.", "date": "2024-10-03", "categories": []}
{"id": "2306.02105", "title": "Advancing African-Accented Speech Recognition: Epistemic Uncertainty-Driven Data Selection for Generalizable ASR Models", "abstract": "Accents play a pivotal role in shaping human communication, enhancing our ability to convey and comprehend messages with clarity and cultural nuance. While there has been significant progress in Automatic Speech Recognition (ASR), African-accented English ASR has been understudied due to a lack of training datasets, which are often expensive to create and demand colossal human labor. Combining several active learning paradigms and the core-set approach, we propose a new multi-rounds adaptation process that uses epistemic uncertainty to automate the annotation process, significantly reducing the associated costs and human labor. This novel method streamlines data annotation and strategically selects data samples contributing most to model uncertainty, enhancing training efficiency. We define a new U-WER metric to track model adaptation to hard accents. We evaluate our approach across several domains, datasets, and high-performing speech models. Our results show that our approach leads to a 27\\% WER relative average improvement while requiring on average 45\\% less data than established baselines. Our approach also improves out-of-distribution generalization for very low-resource accents, demonstrating its viability for building generalizable ASR models in the context of accented African ASR. We open-source the code here: https://github.com/bonaventuredossou/active_learning_african_asr.", "date": "2023-06-03", "categories": []}
{"id": "2306.08954", "title": "An Expanded Benchmark that Rediscovers and Affirms the Edge of Uncertainty Sampling for Active Learning in Tabular Datasets", "abstract": "Active Learning (AL) addresses the crucial challenge of enabling machines to efficiently gather labeled examples through strategic queries. Among the many AL strategies, Uncertainty Sampling (US) stands out as one of the most widely adopted. US queries the example(s) that the current model finds uncertain, proving to be both straightforward and effective. Despite claims in the literature suggesting superior alternatives to US, community-wide acceptance remains elusive. In fact, existing benchmarks for tabular datasets present conflicting conclusions on the continued competitiveness of US. In this study, we review the literature on AL strategies in the last decade and build the most comprehensive open-source AL benchmark to date to understand the relative merits of different AL strategies. The benchmark surpasses existing ones by encompassing a broader coverage of strategies, models, and data. Through our investigation of the conflicting conclusions in existing tabular AL benchmarks by evaluation under broad AL experimental settings, we uncover fresh insights into the often-overlooked issue of using machine learning models--**model compatibility** in the context of US. Specifically, we notice that adopting the different models for the querying unlabeled examples and learning tasks would degrade US's effectiveness. Notably, our findings affirm that US maintains a competitive edge over other strategies when paired with compatible models. These findings have practical implications and provide a concrete recipe for AL practitioners, empowering them to make informed decisions when working with tabular classifications with limited labeled data. The code for this project is available on https://github.com/ariapoy/active-learning-benchmark.", "date": "2023-06-15", "categories": []}
{"id": "2506.15309", "title": "Active Learning-Guided Seq2Seq Variational Autoencoder for Multi-target Inhibitor Generation", "abstract": "Simultaneously optimizing molecules against multiple therapeutic targets remains a profound challenge in drug discovery, particularly due to sparse rewards and conflicting design constraints. We propose a structured active learning (AL) paradigm integrating a sequence-to-sequence (Seq2Seq) variational autoencoder (VAE) into iterative loops designed to balance chemical diversity, molecular quality, and multi-target affinity. Our method alternates between expanding chemically feasible regions of latent space and progressively constraining molecules based on increasingly stringent multi-target docking thresholds. In a proof-of-concept study targeting three related coronavirus main proteases (SARS-CoV-2, SARS-CoV, MERS-CoV), our approach efficiently generated a structurally diverse set of pan-inhibitor candidates. We demonstrate that careful timing and strategic placement of chemical filters within this active learning pipeline markedly enhance exploration of beneficial chemical space, transforming the sparse-reward, multi-objective drug design problem into an accessible computational task. Our framework thus provides a generalizable roadmap for efficiently navigating complex polypharmacological landscapes.", "date": "2025-06-18", "categories": []}
{"id": "2506.17607", "title": "Towards Fundamental Limits for Active Multi-distribution Learning", "abstract": "Multi-distribution learning extends agnostic Probably Approximately Correct (PAC) learning to the setting in which a family of $k$ distributions, $\\{D_i\\}_{i\\in[k]}$, is considered and a classifier's performance is measured by its error under the worst distribution. This problem has attracted a lot of recent interests due to its applications in collaborative learning, fairness, and robustness. Despite a rather complete picture of sample complexity of passive multi-distribution learning, research on active multi-distribution learning remains scarce, with algorithms whose optimality remaining unknown.\n  In this paper, we develop new algorithms for active multi-distribution learning and establish improved label complexity upper and lower bounds, in distribution-dependent and distribution-free settings. Specifically, in the near-realizable setting we prove an upper bound of $\\widetilde{O}\\Bigl(\\theta_{\\max}(d+k)\\ln\\frac{1}{\\varepsilon}\\Bigr)$ and $\\widetilde{O}\\Bigl(\\theta_{\\max}(d+k)\\Bigl(\\ln\\frac{1}{\\varepsilon}+\\frac{\\nu^2}{\\varepsilon^2}\\Bigr)+\\frac{k\\nu}{\\varepsilon^2}\\Bigr)$ in the realizable and agnostic settings respectively, where $\\theta_{\\max}$ is the maximum disagreement coefficient among the $k$ distributions, $d$ is the VC dimension of the hypothesis class, $\\nu$ is the multi-distribution error of the best hypothesis, and $\\varepsilon$ is the target excess error. Moreover, we show that the bound in the realizable setting is information-theoretically optimal and that the $k\\nu/\\varepsilon^2$ term in the agnostic setting is fundamental for proper learners. We also establish instance-dependent sample complexity bound for passive multidistribution learning that smoothly interpolates between realizable and agnostic regimes~\\citep{blum2017collaborative,zhang2024optimal}, which may be of independent interest.", "date": "2025-06-21", "categories": []}
{"id": "2506.08070", "title": "Info-Coevolution: An Efficient Framework for Data Model Coevolution", "abstract": "Machine learning relies heavily on data, yet the continuous growth of real-world data poses challenges for efficient dataset construction and training. A fundamental yet unsolved question is: given our current model and data, does a new data (sample/batch) need annotation/learning? Conventional approaches retain all available data, leading to non-optimal data and training efficiency. Active learning aims to reduce data redundancy by selecting a subset of samples to annotate, while it increases pipeline complexity and introduces bias. In this work, we propose Info-Coevolution, a novel framework that efficiently enables models and data to coevolve through online selective annotation with no bias. Leveraging task-specific models (and open-source models), it selectively annotates and integrates online and web data to improve datasets efficiently. For real-world datasets like ImageNet-1K, Info-Coevolution reduces annotation and training costs by 32\\% without performance loss. It is able to automatically give the saving ratio without tuning the ratio. It can further reduce the annotation ratio to 50\\% with semi-supervised learning. We also explore retrieval-based dataset enhancement using unlabeled open-source data. Code is available at https://github.com/NUS-HPC-AI-Lab/Info-Coevolution/.", "date": "2025-06-09", "categories": []}
{"id": "2506.15893", "title": "Formal Models of Active Learning from Contrastive Examples", "abstract": "Machine learning can greatly benefit from providing learning algorithms with pairs of contrastive training examples -- typically pairs of instances that differ only slightly, yet have different class labels. Intuitively, the difference in the instances helps explain the difference in the class labels. This paper proposes a theoretical framework in which the effect of various types of contrastive examples on active learners is studied formally. The focus is on the sample complexity of learning concept classes and how it is influenced by the choice of contrastive examples. We illustrate our results with geometric concept classes and classes of Boolean functions. Interestingly, we reveal a connection between learning from contrastive examples and the classical model of self-directed learning.", "date": "2025-06-18", "categories": []}
{"id": "2506.13043", "title": "ViewPCL: a point cloud based active learning method for multi-view segmentation", "abstract": "We propose a novel active learning framework for multi-view semantic segmentation. This framework relies on a new score that measures the discrepancy between point cloud distributions generated from the extra geometrical information derived from the model's prediction across different views. Our approach results in a data efficient and explainable active learning method. The source code is available at https://github.com/chilai235/viewpclAL.", "date": "2025-06-16", "categories": []}
{"id": "2506.11100", "title": "An Active Learning-Based Streaming Pipeline for Reduced Data Training of Structure Finding Models in Neutron Diffractometry", "abstract": "Structure determination workloads in neutron diffractometry are computationally expensive and routinely require several hours to many days to determine the structure of a material from its neutron diffraction patterns. The potential for machine learning models trained on simulated neutron scattering patterns to significantly speed up these tasks have been reported recently. However, the amount of simulated data needed to train these models grows exponentially with the number of structural parameters to be predicted and poses a significant computational challenge. To overcome this challenge, we introduce a novel batch-mode active learning (AL) policy that uses uncertainty sampling to simulate training data drawn from a probability distribution that prefers labelled examples about which the model is least certain. We confirm its efficacy in training the same models with about 75% less training data while improving the accuracy. We then discuss the design of an efficient stream-based training workflow that uses this AL policy and present a performance study on two heterogeneous platforms to demonstrate that, compared with a conventional training workflow, the streaming workflow delivers about 20% shorter training time without any loss of accuracy.", "date": "2025-06-06", "categories": []}
{"id": "2312.09196", "title": "Improved Algorithm for Deep Active Learning under Imbalance via Optimal Separation", "abstract": "Class imbalance severely impacts machine learning performance on minority classes in real-world applications. While various solutions exist, active learning offers a fundamental fix by strategically collecting balanced, informative labeled examples from abundant unlabeled data. We introduce DIRECT, an algorithm that identifies class separation boundaries and selects the most uncertain nearby examples for annotation. By reducing the problem to one-dimensional active learning, DIRECT leverages established theory to handle batch labeling and label noise -- another common challenge in data annotation that particularly affects active learning methods. Our work presents the first comprehensive study of active learning under both class imbalance and label noise. Extensive experiments on imbalanced datasets show DIRECT reduces annotation costs by over 60\\% compared to state-of-the-art active learning methods and over 80\\% versus random sampling, while maintaining robustness to label noise.", "date": "2023-12-14", "categories": []}
{"id": "2506.10120", "title": "GRAIL: A Benchmark for GRaph ActIve Learning in Dynamic Sensing Environments", "abstract": "Graph-based Active Learning (AL) leverages the structure of graphs to efficiently prioritize label queries, reducing labeling costs and user burden in applications like health monitoring, human behavior analysis, and sensor networks. By identifying strategically positioned nodes, graph AL minimizes data collection demands while maintaining model performance, making it a valuable tool for dynamic environments. Despite its potential, existing graph AL methods are often evaluated on static graph datasets and primarily focus on prediction accuracy, neglecting user-centric considerations such as sampling diversity, query fairness, and adaptability to dynamic settings. To bridge this gap, we introduce GRAIL, a novel benchmarking framework designed to evaluate graph AL strategies in dynamic, real-world environments. GRAIL introduces novel metrics to assess sustained effectiveness, diversity, and user burden, enabling a comprehensive evaluation of AL methods under varying conditions. Extensive experiments on datasets featuring dynamic, real-life human sensor data reveal trade-offs between prediction performance and user burden, highlighting limitations in existing AL strategies. GRAIL demonstrates the importance of balancing node importance, query diversity, and network topology, providing an evaluation mechanism for graph AL solutions in dynamic environments.", "date": "2025-06-11", "categories": []}
{"id": "2410.11226", "title": "MF-LAL: Drug Compound Generation Using Multi-Fidelity Latent Space Active Learning", "abstract": "Current generative models for drug discovery primarily use molecular docking as an oracle to guide the generation of active compounds. However, such models are often not useful in practice because even compounds with high docking scores do not consistently show real-world experimental activity. More accurate methods for activity prediction exist, such as molecular dynamics based binding free energy calculations, but they are too computationally expensive to use in a generative model. To address this challenge, we propose Multi-Fidelity Latent space Active Learning (MF-LAL), a generative modeling framework that integrates a set of oracles with varying cost-accuracy tradeoffs. Using active learning, we train a surrogate model for each oracle and use these surrogates to guide generation of compounds with high predicted activity. Unlike previous approaches that separately learn the surrogate model and generative model, MF-LAL combines the generative and multi-fidelity surrogate models into a single framework, allowing for more accurate activity prediction and higher quality samples. Our experiments on two disease-relevant proteins show that MF-LAL produces compounds with significantly better binding free energy scores than other single and multi-fidelity approaches (~50% improvement in mean binding free energy score). The code is available at https://github.com/Rose-STL-Lab/MF-LAL.", "date": "2024-10-15", "categories": []}
{"id": "2506.07949", "title": "Cost-Optimal Active AI Model Evaluation", "abstract": "The development lifecycle of generative AI systems requires continual evaluation, data acquisition, and annotation, which is costly in both resources and time. In practice, rapid iteration often makes it necessary to rely on synthetic annotation data because of the low cost, despite the potential for substantial bias. In this paper, we develop novel, cost-aware methods for actively balancing the use of a cheap, but often inaccurate, weak rater -- such as a model-based autorater that is designed to automatically assess the quality of generated content -- with a more expensive, but also more accurate, strong rater alternative such as a human. More specifically, the goal of our approach is to produce a low variance, unbiased estimate of the mean of the target \"strong\" rating, subject to some total annotation budget. Building on recent work in active and prediction-powered statistical inference, we derive a family of cost-optimal policies for allocating a given annotation budget between weak and strong raters so as to maximize statistical efficiency. Using synthetic and real-world data, we empirically characterize the conditions under which these policies yield improvements over prior methods. We find that, especially in tasks where there is high variability in the difficulty of examples, our policies can achieve the same estimation precision at a far lower total annotation budget than standard evaluation methods.", "date": "2025-06-09", "categories": []}
{"id": "2412.20925", "title": "Active Learning with Variational Quantum Circuits for Quantum Process Tomography", "abstract": "Quantum process tomography (QPT) is a fundamental tool for fully characterizing quantum systems. It relies on querying a set of quantum states as input to the quantum process. Previous QPT methods typically employ a straightforward strategy for randomly selecting quantum states, overlooking differences in informativeness among them. In this work, we propose a general active learning (AL) framework that adaptively selects the most informative subset of quantum states for reconstruction. We design and evaluate various AL algorithms and provide practical guidelines for selecting suitable methods in different scenarios. In particular, we introduce a learning framework that leverages the widely-used variational quantum circuits (VQCs) to perform the QPT task and integrate our AL algorithms into the query step. We demonstrate our algorithms by reconstructing the unitary quantum processes resulting from random quantum circuits with up to seven qubits. Numerical results show that our AL algorithms achieve significantly improved reconstruction, and the improvement increases with the size of the underlying quantum system. Our work opens new avenues for further advancing existing QPT methods.", "date": "2024-12-30", "categories": []}
{"id": "2506.00316", "title": "Active Learning via Regression Beyond Realizability", "abstract": "We present a new active learning framework for multiclass classification based on surrogate risk minimization that operates beyond the standard realizability assumption. Existing surrogate-based active learning algorithms crucially rely on realizability$\\unicode{x2014}$the assumption that the optimal surrogate predictor lies within the model class$\\unicode{x2014}$limiting their applicability in practical, misspecified settings. In this work we show that under conditions significantly weaker than realizability, as long as the class of models considered is convex, one can still obtain a label and sample complexity comparable to prior work. Despite achieving similar rates, the algorithmic approaches from prior works can be shown to fail in non-realizable settings where our assumption is satisfied. Our epoch-based active learning algorithm departs from prior methods by fitting a model from the full class to the queried data in each epoch and returning an improper classifier obtained by aggregating these models.", "date": "2025-05-31", "categories": []}
{"id": "2506.03307", "title": "Budgeted Online Active Learning with Expert Advice and Episodic Priors", "abstract": "This paper introduces a novel approach to budgeted online active learning from finite-horizon data streams with extremely limited labeling budgets. In agricultural applications, such streams might include daily weather data over a growing season, and labels require costly measurements of weather-dependent plant characteristics. Our method integrates two key sources of prior information: a collection of preexisting expert predictors and episodic behavioral knowledge of the experts based on unlabeled data streams. Unlike previous research on online active learning with experts, our work simultaneously considers query budgets, finite horizons, and episodic knowledge, enabling effective learning in applications with severely limited labeling capacity. We demonstrate the utility of our approach through experiments on various prediction problems derived from both a realistic agricultural crop simulator and real-world data from multiple grape cultivars. The results show that our method significantly outperforms baseline expert predictions, uniform query selection, and existing approaches that consider budgets and limited horizons but neglect episodic knowledge, even under highly constrained labeling budgets.", "date": "2025-06-03", "categories": []}
{"id": "2506.03817", "title": "Survey of Active Learning Hyperparameters: Insights from a Large-Scale Experimental Grid", "abstract": "Annotating data is a time-consuming and costly task, but it is inherently required for supervised machine learning. Active Learning (AL) is an established method that minimizes human labeling effort by iteratively selecting the most informative unlabeled samples for expert annotation, thereby improving the overall classification performance. Even though AL has been known for decades, AL is still rarely used in real-world applications. As indicated in the two community web surveys among the NLP community about AL, two main reasons continue to hold practitioners back from using AL: first, the complexity of setting AL up, and second, a lack of trust in its effectiveness. We hypothesize that both reasons share the same culprit: the large hyperparameter space of AL. This mostly unexplored hyperparameter space often leads to misleading and irreproducible AL experiment results. In this study, we first compiled a large hyperparameter grid of over 4.6 million hyperparameter combinations, second, recorded the performance of all combinations in the so-far biggest conducted AL study, and third, analyzed the impact of each hyperparameter in the experiment results. In the end, we give recommendations about the influence of each hyperparameter, demonstrate the surprising influence of the concrete AL strategy implementation, and outline an experimental study design for reproducible AL experiments with minimal computational effort, thus contributing to more reproducible and trustworthy AL research in the future.", "date": "2025-06-04", "categories": []}
{"id": "2505.18924", "title": "LLM-Guided Taxonomy and Hierarchical Uncertainty for 3D Point Cloud Active Learning", "abstract": "We present a novel active learning framework for 3D point cloud semantic segmentation that, for the first time, integrates large language models (LLMs) to construct hierarchical label structures and guide uncertainty-based sample selection. Unlike prior methods that treat labels as flat and independent, our approach leverages LLM prompting to automatically generate multi-level semantic taxonomies and introduces a recursive uncertainty projection mechanism that propagates uncertainty across hierarchy levels. This enables spatially diverse, label-aware point selection that respects the inherent semantic structure of 3D scenes. Experiments on S3DIS and ScanNet v2 show that our method achieves up to 4% mIoU improvement under extremely low annotation budgets (e.g., 0.02%), substantially outperforming existing baselines. Our results highlight the untapped potential of LLMs as knowledge priors in 3D vision and establish hierarchical uncertainty modeling as a powerful paradigm for efficient point cloud annotation.", "date": "2025-05-25", "categories": []}
{"id": "2506.01992", "title": "No Free Lunch in Active Learning: LLM Embedding Quality Dictates Query Strategy Success", "abstract": "The advent of large language models (LLMs) capable of producing general-purpose representations lets us revisit the practicality of deep active learning (AL): By leveraging frozen LLM embeddings, we can mitigate the computational costs of iteratively fine-tuning large backbones. This study establishes a benchmark and systematically investigates the influence of LLM embedding quality on query strategies in deep AL. We employ five top-performing models from the massive text embedding benchmark (MTEB) leaderboard and two baselines for ten diverse text classification tasks. Our findings reveal key insights: First, initializing the labeled pool using diversity-based sampling synergizes with high-quality embeddings, boosting performance in early AL iterations. Second, the choice of the optimal query strategy is sensitive to embedding quality. While the computationally inexpensive Margin sampling can achieve performance spikes on specific datasets, we find that strategies like Badge exhibit greater robustness across tasks. Importantly, their effectiveness is often enhanced when paired with higher-quality embeddings. Our results emphasize the need for context-specific evaluation of AL strategies, as performance heavily depends on embedding quality and the target task.", "date": "2025-05-18", "categories": []}
{"id": "2506.02753", "title": "Multi-task Learning with Active Learning for Arabic Offensive Speech Detection", "abstract": "The rapid growth of social media has amplified the spread of offensive, violent, and vulgar speech, which poses serious societal and cybersecurity concerns. Detecting such content in Arabic text is particularly complex due to limited labeled data, dialectal variations, and the language's inherent complexity. This paper proposes a novel framework that integrates multi-task learning (MTL) with active learning to enhance offensive speech detection in Arabic social media text. By jointly training on two auxiliary tasks, violent and vulgar speech, the model leverages shared representations to improve the detection accuracy of the offensive speech. Our approach dynamically adjusts task weights during training to balance the contribution of each task and optimize performance. To address the scarcity of labeled data, we employ an active learning strategy through several uncertainty sampling techniques to iteratively select the most informative samples for model training. We also introduce weighted emoji handling to better capture semantic cues. Experimental results on the OSACT2022 dataset show that the proposed framework achieves a state-of-the-art macro F1-score of 85.42%, outperforming existing methods while using significantly fewer fine-tuning samples. The findings of this study highlight the potential of integrating MTL with active learning for efficient and accurate offensive language detection in resource-constrained settings.", "date": "2025-06-03", "categories": []}
{"id": "2408.03819", "title": "Leveraging Variation Theory in Counterfactual Data Augmentation for Optimized Active Learning", "abstract": "Active Learning (AL) allows models to learn interactively from user feedback. This paper introduces a counterfactual data augmentation approach to AL, particularly addressing the selection of datapoints for user querying, a pivotal concern in enhancing data efficiency. Our approach is inspired by Variation Theory, a theory of human concept learning that emphasizes the essential features of a concept by focusing on what stays the same and what changes. Instead of just querying with existing datapoints, our approach synthesizes artificial datapoints that highlight potential key similarities and differences among labels using a neuro-symbolic pipeline combining large language models (LLMs) and rule-based models. Through an experiment in the example domain of text classification, we show that our approach achieves significantly higher performance when there are fewer annotated data. As the annotated training data gets larger the impact of the generated data starts to diminish showing its capability to address the cold start problem in AL. This research sheds light on integrating theories of human learning into the optimization of AL.", "date": "2024-08-07", "categories": []}
{"id": "2411.00524", "title": "Comparison-based Active Preference Learning for Multi-dimensional Personalization", "abstract": "Large language models (LLMs) have shown remarkable success, but aligning them with human preferences remains a core challenge. As individuals have their own, multi-dimensional preferences, recent studies have explored multi-dimensional personalization, which aims to enable models to generate responses personalized to explicit preferences. However, human preferences are often implicit and thus difficult to articulate, limiting the direct application of this approach. To bridge this gap, we propose Active Multi-dimensional Preference Learning (AMPLe), designed to capture implicit user preferences from interactively collected comparative feedback. Building on Bayesian inference, our work introduces a modified posterior update procedure to mitigate estimation bias and potential noise in comparisons. Also, inspired by generalized binary search, we employ an active query selection strategy to minimize the number of required comparisons by a user. Through theoretical analysis and experiments on language generation tasks, we demonstrate feedback efficiency and effectiveness of our framework in personalizing model responses. Our code is publicly available at https://github.com/ml-postech/AMPLe .", "date": "2024-11-01", "categories": []}
{"id": "2502.11767", "title": "From Selection to Generation: A Survey of LLM-based Active Learning", "abstract": "Active Learning (AL) has been a powerful paradigm for improving model efficiency and performance by selecting the most informative data points for labeling and training. In recent active learning frameworks, Large Language Models (LLMs) have been employed not only for selection but also for generating entirely new data instances and providing more cost-effective annotations. Motivated by the increasing importance of high-quality data and efficient model training in the era of LLMs, we present a comprehensive survey on LLM-based Active Learning. We introduce an intuitive taxonomy that categorizes these techniques and discuss the transformative roles LLMs can play in the active learning loop. We further examine the impact of AL on LLM learning paradigms and its applications across various domains. Finally, we identify open challenges and propose future research directions. This survey aims to serve as an up-to-date resource for researchers and practitioners seeking to gain an intuitive understanding of LLM-based AL techniques and deploy them to new applications.", "date": "2025-02-17", "categories": []}
{"id": "2506.01724", "title": "Active Learning via Vision-Language Model Adaptation with Open Data", "abstract": "Pretrained on web-scale open data, VLMs offer powerful capabilities for solving downstream tasks after being adapted to task-specific labeled data. Yet, data labeling can be expensive and may demand domain expertise. Active Learning (AL) aims to reduce this expense by strategically selecting the most informative data for labeling and model training. Recent AL methods have explored VLMs but have not leveraged publicly available open data, such as VLM's pretraining data. In this work, we leverage such data by retrieving task-relevant examples to augment the task-specific examples. As expected, incorporating them significantly improves AL. Given that our method exploits open-source VLM and open data, we refer to it as Active Learning with Open Resources (ALOR). Additionally, most VLM-based AL methods use prompt tuning (PT) for model adaptation, likely due to its ability to directly utilize pretrained parameters and the assumption that doing so reduces the risk of overfitting to limited labeled data. We rigorously compare popular adaptation approaches, including linear probing (LP), finetuning (FT), and contrastive tuning (CT). We reveal two key findings: (1) All adaptation approaches benefit from incorporating retrieved data, and (2) CT resoundingly outperforms other approaches across AL methods. Further analysis of retrieved data reveals a naturally imbalanced distribution of task-relevant classes, exposing inherent biases within the VLM. This motivates our novel Tail First Sampling (TFS) strategy for AL, an embarrassingly simple yet effective method that prioritizes sampling data from underrepresented classes to label. Extensive experiments demonstrate that our final method, contrastively finetuning VLM on both retrieved and TFS-selected labeled data, significantly outperforms existing methods.", "date": "2025-06-02", "categories": []}
{"id": "2505.20956", "title": "Hybrid Disagreement-Diversity Active Learning for Bioacoustic Sound Event Detection", "abstract": "Bioacoustic sound event detection (BioSED) is crucial for biodiversity conservation but faces practical challenges during model development and training: limited amounts of annotated data, sparse events, species diversity, and class imbalance. To address these challenges efficiently with a limited labeling budget, we apply the mismatch-first farthest-traversal (MFFT), an active learning method integrating committee voting disagreement and diversity analysis. We also refine an existing BioSED dataset specifically for evaluating active learning algorithms. Experimental results demonstrate that MFFT achieves a mAP of 68% when cold-starting and 71% when warm-starting (which is close to the fully-supervised mAP of 75%) while using only 2.3% of the annotations. Notably, MFFT excels in cold-start scenarios and with rare species, which are critical for monitoring endangered species, demonstrating its practical value.", "date": "2025-05-27", "categories": []}
{"id": "2505.20691", "title": "Evidential Deep Active Learning for Semi-Supervised Classification", "abstract": "Semi-supervised classification based on active learning has made significant progress, but the existing methods often ignore the uncertainty estimation (or reliability) of the prediction results during the learning process, which makes it questionable whether the selected samples can effectively update the model. Hence, this paper proposes an evidential deep active learning approach for semi-supervised classification (EDALSSC). EDALSSC builds a semi-supervised learning framework to simultaneously quantify the uncertainty estimation of labeled and unlabeled data during the learning process. The uncertainty estimation of the former is associated with evidential deep learning, while that of the latter is modeled by combining ignorance information and conflict information of the evidence from the perspective of the T-conorm operator. Furthermore, this article constructs a heuristic method to dynamically balance the influence of evidence and the number of classes on uncertainty estimation to ensure that it does not produce counter-intuitive results in EDALSSC. For the sample selection strategy, EDALSSC selects the sample with the greatest uncertainty estimation that is calculated in the form of a sum when the training loss increases in the latter half of the learning process. Experimental results demonstrate that EDALSSC outperforms existing semi-supervised and supervised active learning approaches on image classification datasets.", "date": "2025-05-27", "categories": []}
{"id": "2307.10434", "title": "Learning Formal Specifications from Membership and Preference Queries", "abstract": "Active learning is a well-studied approach to learning formal specifications, such as automata. In this work, we extend active specification learning by proposing a novel framework that strategically requests a combination of membership labels and pair-wise preferences, a popular alternative to membership labels. The combination of pair-wise preferences and membership labels allows for a more flexible approach to active specification learning, which previously relied on membership labels only. We instantiate our framework in two different domains, demonstrating the generality of our approach. Our results suggest that learning from both modalities allows us to robustly and conveniently identify specifications via membership and preferences.", "date": "2023-07-19", "categories": []}
{"id": "2505.14882", "title": "An active learning framework for multi-group mean estimation", "abstract": "We study a fundamental learning problem over multiple groups with unknown data distributions, where an analyst would like to learn the mean of each group. Moreover, we want to ensure that this data is collected in a relatively fair manner such that the noise of the estimate of each group is reasonable. In particular, we focus on settings where data are collected dynamically, which is important in adaptive experimentation for online platforms or adaptive clinical trials for healthcare. In our model, we employ an active learning framework to sequentially collect samples with bandit feedback, observing a sample in each period from the chosen group. After observing a sample, the analyst updates their estimate of the mean and variance of that group and chooses the next group accordingly. The analyst's objective is to dynamically collect samples to minimize the collective noise of the estimators, measured by the norm of the vector of variances of the mean estimators.\n  We propose an algorithm, Variance-UCB, that sequentially selects groups according to an upper confidence bound on the variance estimate. We provide a general theoretical framework for providing efficient bounds on learning from any underlying distribution where the variances can be estimated reasonably. This framework yields upper bounds on regret that improve significantly upon all existing bounds, as well as a collection of new results for different objectives and distributions than those previously studied.", "date": "2025-05-20", "categories": []}
{"id": "2505.14964", "title": "The Achilles Heel of AI: Fundamentals of Risk-Aware Training Data for High-Consequence Models", "abstract": "AI systems in high-consequence domains such as defense, intelligence, and disaster response must detect rare, high-impact events while operating under tight resource constraints. Traditional annotation strategies that prioritize label volume over informational value introduce redundancy and noise, limiting model generalization. This paper introduces smart-sizing, a training data strategy that emphasizes label diversity, model-guided selection, and marginal utility-based stopping. We implement this through Adaptive Label Optimization (ALO), combining pre-labeling triage, annotator disagreement analysis, and iterative feedback to prioritize labels that meaningfully improve model performance. Experiments show that models trained on 20 to 40 percent of curated data can match or exceed full-data baselines, particularly in rare-class recall and edge-case generalization. We also demonstrate how latent labeling errors embedded in training and validation sets can distort evaluation, underscoring the need for embedded audit tools and performance-aware governance. Smart-sizing reframes annotation as a feedback-driven process aligned with mission outcomes, enabling more robust models with fewer labels and supporting efficient AI development pipelines for frontier models and operational systems.", "date": "2025-05-20", "categories": []}
{"id": "2410.13853", "title": "AutoAL: Automated Active Learning with Differentiable Query Strategy Search", "abstract": "As deep learning continues to evolve, the need for data efficiency becomes increasingly important. Considering labeling large datasets is both time-consuming and expensive, active learning (AL) provides a promising solution to this challenge by iteratively selecting the most informative subsets of examples to train deep neural networks, thereby reducing the labeling cost. However, the effectiveness of different AL algorithms can vary significantly across data scenarios, and determining which AL algorithm best fits a given task remains a challenging problem. This work presents the first differentiable AL strategy search method, named AutoAL, which is designed on top of existing AL sampling strategies. AutoAL consists of two neural nets, named SearchNet and FitNet, which are optimized concurrently under a differentiable bi-level optimization framework. For any given task, SearchNet and FitNet are iteratively co-optimized using the labeled data, learning how well a set of candidate AL algorithms perform on that task. With the optimal AL strategies identified, SearchNet selects a small subset from the unlabeled pool for querying their annotations, enabling efficient training of the task model. Experimental results demonstrate that AutoAL consistently achieves superior accuracy compared to all candidate AL algorithms and other selective AL approaches, showcasing its potential for adapting and integrating multiple existing AL methods across diverse tasks and domains. Code is available at: https://github.com/haizailache999/AutoAL.", "date": "2024-10-17", "categories": []}
{"id": "2502.08200", "title": "ActiveSSF: An Active-Learning-Guided Self-Supervised Framework for Long-Tailed Megakaryocyte Classification", "abstract": "Precise classification of megakaryocytes is crucial for diagnosing myelodysplastic syndromes. Although self-supervised learning has shown promise in medical image analysis, its application to classifying megakaryocytes in stained slides faces three main challenges: (1) pervasive background noise that obscures cellular details, (2) a long-tailed distribution that limits data for rare subtypes, and (3) complex morphological variations leading to high intra-class variability. To address these issues, we propose the ActiveSSF framework, which integrates active learning with self-supervised pretraining. Specifically, our approach employs Gaussian filtering combined with K-means clustering and HSV analysis (augmented by clinical prior knowledge) for accurate region-of-interest extraction; an adaptive sample selection mechanism that dynamically adjusts similarity thresholds to mitigate class imbalance; and prototype clustering on labeled samples to overcome morphological complexity. Experimental results on clinical megakaryocyte datasets demonstrate that ActiveSSF not only achieves state-of-the-art performance but also significantly improves recognition accuracy for rare subtypes. Moreover, the integration of these advanced techniques further underscores the practical potential of ActiveSSF in clinical settings.", "date": "2025-02-12", "categories": []}
{"id": "2305.04539", "title": "Q&A Label Learning", "abstract": "Assigning labels to instances is crucial for supervised machine learning. In this paper, we proposed a novel annotation method called Q&A labeling, which involves a question generator that asks questions about the labels of the instances to be assigned, and an annotator who answers the questions and assigns the corresponding labels to the instances. We derived a generative model of labels assigned according to two different Q&A labeling procedures that differ in the way questions are asked and answered. We showed that, in both procedures, the derived model is partially consistent with that assumed in previous studies. The main distinction of this study from previous studies lies in the fact that the label generative model was not assumed, but rather derived based on the definition of a specific annotation method, Q&A labeling. We also derived a loss function to evaluate the classification risk of ordinary supervised machine learning using instances assigned Q&A labels and evaluated the upper bound of the classification error. The results indicate statistical consistency in learning with Q&A labels.", "date": "2023-05-08", "categories": []}
{"id": "2504.03783", "title": "FAST: Federated Active Learning with Foundation Models for Communication-efficient Sampling and Training", "abstract": "Federated Active Learning (FAL) has emerged as a promising framework to leverage large quantities of unlabeled data across distributed clients while preserving data privacy. However, real-world deployments remain limited by high annotation costs and communication-intensive sampling processes, particularly in a cross-silo setting, when clients possess substantial local datasets. This paper addresses the crucial question: What is the best practice to reduce communication costs in human-in-the-loop learning with minimal annotator effort? Existing FAL methods typically rely on iterative annotation processes that separate active sampling from federated updates, leading to multiple rounds of expensive communication and annotation. In response, we introduce FAST, a two-pass FAL framework that harnesses foundation models for weak labeling in a preliminary pass, followed by a refinement pass focused exclusively on the most uncertain samples. By leveraging representation knowledge from foundation models and integrating refinement steps into a streamlined workflow, FAST substantially reduces the overhead incurred by iterative active sampling. Extensive experiments on diverse medical and natural image benchmarks demonstrate that FAST outperforms existing FAL methods by an average of 4.36% while reducing communication rounds eightfold under a limited 5% labeling budget.", "date": "2025-04-03", "categories": []}
{"id": "2505.12913", "title": "Active Learning on Synthons for Molecular Design", "abstract": "Exhaustive virtual screening is highly informative but often intractable against the expensive objective functions involved in modern drug discovery. This problem is exacerbated in combinatorial contexts such as multi-vector expansion, where molecular spaces can quickly become ultra-large. Here, we introduce Scalable Active Learning via Synthon Acquisition (SALSA): a simple algorithm applicable to multi-vector expansion which extends pool-based active learning to non-enumerable spaces by factoring modeling and acquisition over synthon or fragment choices. Through experiments on ligand- and structure-based objectives, we highlight SALSA's sample efficiency, and its ability to scale to spaces of trillions of compounds. Further, we demonstrate application toward multi-parameter objective design tasks on three protein targets - finding SALSA-generated molecules have comparable chemical property profiles to known bioactives, and exhibit greater diversity and higher scores over an industry-leading generative approach.", "date": "2025-05-19", "categories": []}
{"id": "2505.06825", "title": "Active Learning for Multi-class Image Classification", "abstract": "A principle bottleneck in image classification is the large number of training examples needed to train a classifier. Using active learning, we can reduce the number of training examples to teach a CNN classifier by strategically selecting examples. Assigning values to image examples using different uncertainty metrics allows the model to identify and select high-value examples in a smaller training set size. We demonstrate results for digit recognition and fruit classification on the MNIST and Fruits360 data sets. We formally compare results for four different uncertainty metrics. Finally, we observe active learning is also effective on simpler (binary) classification tasks, but marked improvement from random sampling is more evident on more difficult tasks. We show active learning is a viable algorithm for image classification problems.", "date": "2025-05-11", "categories": []}
{"id": "2412.02529", "title": "Active learning of neural population dynamics using two-photon\n  holographic optogenetics", "abstract": "  Recent advances in techniques for monitoring and perturbing neural\npopulations have greatly enhanced our ability to study circuits in the brain.\nIn particular, two-photon holographic optogenetics now enables precise\nphotostimulation of experimenter-specified groups of individual neurons, while\nsimultaneous two-photon calcium imaging enables the measurement of ongoing and\ninduced activity across the neural population. Despite the enormous space of\npotential photostimulation patterns and the time-consuming nature of\nphotostimulation experiments, very little algorithmic work has been done to\ndetermine the most effective photostimulation patterns for identifying the\nneural population dynamics. Here, we develop methods to efficiently select\nwhich neurons to stimulate such that the resulting neural responses will best\ninform a dynamical model of the neural population activity. Using neural\npopulation responses to photostimulation in mouse motor cortex, we demonstrate\nthe efficacy of a low-rank linear dynamical systems model, and develop an\nactive learning procedure which takes advantage of low-rank structure to\ndetermine informative photostimulation patterns. We demonstrate our approach on\nboth real and synthetic data, obtaining in some cases as much as a two-fold\nreduction in the amount of data required to reach a given predictive power. Our\nactive stimulation design method is based on a novel active learning procedure\nfor low-rank regression, which may be of independent interest.\n", "date": "2024-12-03", "categories": []}
{"id": "2505.04376", "title": "Label-efficient Single Photon Images Classification via Active Learning", "abstract": "  Single-photon LiDAR achieves high-precision 3D imaging in extreme\nenvironments through quantum-level photon detection technology. Current\nresearch primarily focuses on reconstructing 3D scenes from sparse photon\nevents, whereas the semantic interpretation of single-photon images remains\nunderexplored, due to high annotation costs and inefficient labeling\nstrategies. This paper presents the first active learning framework for\nsingle-photon image classification. The core contribution is an imaging\ncondition-aware sampling strategy that integrates synthetic augmentation to\nmodel variability across imaging conditions. By identifying samples where the\nmodel is both uncertain and sensitive to these conditions, the proposed method\nselectively annotates only the most informative examples. Experiments on both\nsynthetic and real-world datasets show that our approach outperforms all\nbaselines and achieves high classification accuracy with significantly fewer\nlabeled samples. Specifically, our approach achieves 97% accuracy on synthetic\nsingle-photon data using only 1.5% labeled samples. On real-world data, we\nmaintain 90.63% accuracy with just 8% labeled samples, which is 4.51% higher\nthan the best-performing baseline. This illustrates that active learning\nenables the same level of classification performance on single-photon images as\non classical images, opening doors to large-scale integration of single-photon\ndata in real-world applications.\n", "date": "2025-05-07", "categories": []}
{"id": "2505.05224", "title": "GFlowNets for Active Learning Based Resource Allocation in Next\n  Generation Wireless Networks", "abstract": "  In this work, we consider the radio resource allocation problem in a wireless\nsystem with various integrated functionalities, such as communication, sensing\nand computing. We design suitable resource management techniques that can\nsimultaneously cater to those heterogeneous requirements, and scale\nappropriately with the high-dimensional and discrete nature of the problem. We\npropose a novel active learning framework where resource allocation patterns\nare drawn sequentially, evaluated in the environment, and then used to\niteratively update a surrogate model of the environment. Our method leverages a\ngenerative flow network (GFlowNet) to sample favorable solutions, as such\nmodels are trained to generate compositional objects proportionally to their\ntraining reward, hence providing an appropriate coverage of its modes. As such,\nGFlowNet generates diverse and high return resource management designs that\nupdate the surrogate model and swiftly discover suitable solutions. We provide\nsimulation results showing that our method can allocate radio resources\nachieving 20% performance gains against benchmarks, while requiring less than\nhalf of the number of acquisition rounds.\n", "date": "2025-05-08", "categories": []}
{"id": "2505.04375", "title": "Balancing Accuracy, Calibration, and Efficiency in Active Learning with\n  Vision Transformers Under Label Noise", "abstract": "  Fine-tuning pre-trained convolutional neural networks on ImageNet for\ndownstream tasks is well-established. Still, the impact of model size on the\nperformance of vision transformers in similar scenarios, particularly under\nlabel noise, remains largely unexplored. Given the utility and versatility of\ntransformer architectures, this study investigates their practicality under\nlow-budget constraints and noisy labels. We explore how classification accuracy\nand calibration are affected by symmetric label noise in active learning\nsettings, evaluating four vision transformer configurations (Base and Large\nwith 16x16 and 32x32 patch sizes) and three Swin Transformer configurations\n(Tiny, Small, and Base) on CIFAR10 and CIFAR100 datasets, under varying label\nnoise rates. Our findings show that larger ViT models (ViTl32 in particular)\nconsistently outperform their smaller counterparts in both accuracy and\ncalibration, even under moderate to high label noise, while Swin Transformers\nexhibit weaker robustness across all noise levels. We find that smaller patch\nsizes do not always lead to better performance, as ViTl16 performs consistently\nworse than ViTl32 while incurring a higher computational cost. We also find\nthat information-based Active Learning strategies only provide meaningful\naccuracy improvements at moderate label noise rates, but they result in poorer\ncalibration compared to models trained on randomly acquired labels, especially\nat high label noise rates. We hope these insights provide actionable guidance\nfor practitioners looking to deploy vision transformers in resource-constrained\nenvironments, where balancing model complexity, label noise, and compute\nefficiency is critical in model fine-tuning or distillation.\n", "date": "2025-05-07", "categories": []}
{"id": "2310.08387", "title": "Aligning Data Selection with Performance: Performance-driven\n  Reinforcement Learning for Active Learning in Object Detection", "abstract": "  Active learning strategies aim to train high-performance models with minimal\nlabeled data by selecting the most informative instances for labeling. However,\nexisting methods for assessing data informativeness often fail to align\ndirectly with task model performance metrics, such as mean average precision\n(mAP) in object detection. This paper introduces Mean-AP Guided Reinforced\nActive Learning for Object Detection (MGRAL), a novel approach that leverages\nthe concept of expected model output changes as informativeness for deep\ndetection networks, directly optimizing the sampling strategy using mAP. MGRAL\nemploys a reinforcement learning agent based on LSTM architecture to\nefficiently navigate the combinatorial challenge of batch sample selection and\nthe non-differentiable nature between performance and selected batches. The\nagent optimizes selection using policy gradient with mAP improvement as the\nreward signal. To address the computational intensity of mAP estimation with\nunlabeled samples, we implement fast look-up tables, ensuring real-world\nfeasibility. We evaluate MGRAL on PASCAL VOC and MS COCO benchmarks across\nvarious backbone architectures. Our approach demonstrates strong performance,\nestablishing a new paradigm in reinforcement learning-based active learning for\nobject detection.\n", "date": "2023-10-12", "categories": []}
{"id": "2505.00507", "title": "HeAL3D: Heuristical-enhanced Active Learning for 3D Object Detection", "abstract": "  Active Learning has proved to be a relevant approach to perform sample\nselection for training models for Autonomous Driving. Particularly, previous\nworks on active learning for 3D object detection have shown that selection of\nsamples in uncontrolled scenarios is challenging. Furthermore, current\napproaches focus exclusively on the theoretical aspects of the sample selection\nproblem but neglect the practical insights that can be obtained from the\nextensive literature and application of 3D detection models. In this paper, we\nintroduce HeAL (Heuristical-enhanced Active Learning for 3D Object Detection)\nwhich integrates those heuristical features together with Localization and\nClassification to deliver the most contributing samples to the model's\ntraining. In contrast to previous works, our approach integrates heuristical\nfeatures such as object distance and point-quantity to estimate the\nuncertainty, which enhance the usefulness of selected samples to train\ndetection models. Our quantitative evaluation on KITTI shows that HeAL presents\ncompetitive mAP with respect to the State-of-the-Art, and achieves the same mAP\nas the full-supervised baseline with only 24% of the samples.\n", "date": "2025-05-01", "categories": []}
{"id": "2502.16892", "title": "Applying LLMs to Active Learning: Towards Cost-Efficient Cross-Task Text\n  Classification without Manually Labeled Data", "abstract": "  Machine learning-based classifiers have been used for text classification,\nsuch as sentiment analysis, news classification, and toxic comment\nclassification. However, supervised machine learning models often require large\namounts of labeled data for training, and manual annotation is both\nlabor-intensive and requires domain-specific knowledge, leading to relatively\nhigh annotation costs. To address this issue, we propose an approach that\nintegrates large language models (LLMs) into an active learning framework,\nachieving high cross-task text classification performance without the need for\nany manually labeled data. Furthermore, compared to directly applying GPT for\nclassification tasks, our approach retains over 93% of its classification\nperformance while requiring only approximately 6% of the computational time and\nmonetary cost, effectively balancing performance and resource efficiency. These\nfindings provide new insights into the efficient utilization of LLMs and active\nlearning algorithms in text classification tasks, paving the way for their\nbroader application.\n", "date": "2025-02-24", "categories": []}
{"id": "2504.19012", "title": "Geometry-aware Active Learning of Spatiotemporal Dynamic Systems", "abstract": "  Rapid developments in advanced sensing and imaging have significantly\nenhanced information visibility, opening opportunities for predictive modeling\nof complex dynamic systems. However, sensing signals acquired from such complex\nsystems are often distributed across 3D geometries and rapidly evolving over\ntime, posing significant challenges in spatiotemporal predictive modeling. This\npaper proposes a geometry-aware active learning framework for modeling\nspatiotemporal dynamic systems. Specifically, we propose a geometry-aware\nspatiotemporal Gaussian Process (G-ST-GP) to effectively integrate the temporal\ncorrelations and geometric manifold features for reliable prediction of\nhigh-dimensional dynamic behaviors. In addition, we develop an adaptive active\nlearning strategy to strategically identify informative spatial locations for\ndata collection and further maximize the prediction accuracy. This strategy\nachieves the adaptive trade-off between the prediction uncertainty in the\nG-ST-GP model and the space-filling design guided by the geodesic distance\nacross the 3D geometry. We implement the proposed framework to model the\nspatiotemporal electrodynamics in a 3D heart geometry. Numerical experiments\nshow that our framework outperforms traditional methods lacking the mechanism\nof geometric information incorporation or effective data collection.\n", "date": "2025-04-26", "categories": []}
{"id": "2505.00511", "title": "Inconsistency-based Active Learning for LiDAR Object Detection", "abstract": "  Deep learning models for object detection in autonomous driving have recently\nachieved impressive performance gains and are already being deployed in\nvehicles worldwide. However, current models require increasingly large datasets\nfor training. Acquiring and labeling such data is costly, necessitating the\ndevelopment of new strategies to optimize this process. Active learning is a\npromising approach that has been extensively researched in the image domain. In\nour work, we extend this concept to the LiDAR domain by developing several\ninconsistency-based sample selection strategies and evaluate their\neffectiveness in various settings. Our results show that using a naive\ninconsistency approach based on the number of detected boxes, we achieve the\nsame mAP as the random sampling strategy with 50% of the labeled data.\n", "date": "2025-05-01", "categories": []}
{"id": "2504.18666", "title": "Co-Training with Active Contrastive Learning and Meta-Pseudo-Labeling on\n  2D Projections for Deep Semi-Supervised Learning", "abstract": "  A major challenge that prevents the training of DL models is the limited\navailability of accurately labeled data. This shortcoming is highlighted in\nareas where data annotation becomes a time-consuming and error-prone task. In\nthis regard, SSL tackles this challenge by capitalizing on scarce labeled and\nabundant unlabeled data; however, SoTA methods typically depend on pre-trained\nfeatures and large validation sets to learn effective representations for\nclassification tasks. In addition, the reduced set of labeled data is often\nrandomly sampled, neglecting the selection of more informative samples. Here,\nwe present active-DeepFA, a method that effectively combines CL,\nteacher-student-based meta-pseudo-labeling and AL to train non-pretrained CNN\narchitectures for image classification in scenarios of scarcity of labeled and\nabundance of unlabeled data. It integrates DeepFA into a co-training setup that\nimplements two cooperative networks to mitigate confirmation bias from\npseudo-labels. The method starts with a reduced set of labeled samples by\nwarming up the networks with supervised CL. Afterward and at regular epoch\nintervals, label propagation is performed on the 2D projections of the\nnetworks' deep features. Next, the most reliable pseudo-labels are exchanged\nbetween networks in a cross-training fashion, while the most meaningful samples\nare annotated and added into the labeled set. The networks independently\nminimize an objective loss function comprising supervised contrastive,\nsupervised and semi-supervised loss components, enhancing the representations\ntowards image classification. Our approach is evaluated on three challenging\nbiological image datasets using only 5% of labeled samples, improving baselines\nand outperforming six other SoTA methods. In addition, it reduces annotation\neffort by achieving comparable results to those of its counterparts with only\n3% of labeled data.\n", "date": "2025-04-25", "categories": []}
{"id": "2504.19014", "title": "Towards minimax optimal algorithms for Active Simple Hypothesis Testing", "abstract": "  We study the Active Simple Hypothesis Testing (ASHT) problem, a simpler\nvariant of the Fixed Budget Best Arm Identification problem. In this work, we\nprovide novel game theoretic formulation of the upper bounds of the ASHT\nproblem. This formulation allows us to leverage tools of differential games and\nPartial Differential Equations (PDEs) to propose an approximately optimal\nalgorithm that is computationally tractable compared to prior work. However,\nthe optimal algorithm still suffers from a curse of dimensionality and instead\nwe use a novel link to Blackwell Approachability to propose an algorithm that\nis far more efficient computationally. We show that this new algorithm,\nalthough not proven to be optimal, is always better than static algorithms in\nall instances of ASHT and is numerically observed to attain the optimal\nexponent in various instances.\n", "date": "2025-04-26", "categories": []}
{"id": "2504.17448", "title": "CHASe: Client Heterogeneity-Aware Data Selection for Effective Federated\n  Active Learning", "abstract": "  Active learning (AL) reduces human annotation costs for machine learning\nsystems by strategically selecting the most informative unlabeled data for\nannotation, but performing it individually may still be insufficient due to\nrestricted data diversity and annotation budget. Federated Active Learning\n(FAL) addresses this by facilitating collaborative data selection and model\ntraining, while preserving the confidentiality of raw data samples. Yet,\nexisting FAL methods fail to account for the heterogeneity of data distribution\nacross clients and the associated fluctuations in global and local model\nparameters, adversely affecting model accuracy. To overcome these challenges,\nwe propose CHASe (Client Heterogeneity-Aware Data Selection), specifically\ndesigned for FAL. CHASe focuses on identifying those unlabeled samples with\nhigh epistemic variations (EVs), which notably oscillate around the decision\nboundaries during training. To achieve both effectiveness and efficiency,\n\\model{} encompasses techniques for 1) tracking EVs by analyzing inference\ninconsistencies across training epochs, 2) calibrating decision boundaries of\ninaccurate models with a new alignment loss, and 3) enhancing data selection\nefficiency via a data freeze and awaken mechanism with subset sampling.\nExperiments show that CHASe surpasses various established baselines in terms of\neffectiveness and efficiency, validated across diverse datasets, model\ncomplexities, and heterogeneous federation settings.\n", "date": "2025-04-24", "categories": []}
{"id": "2406.14388", "title": "Active Diffusion Subsampling", "abstract": "  Subsampling is commonly used to mitigate costs associated with data\nacquisition, such as time or energy requirements, motivating the development of\nalgorithms for estimating the fully-sampled signal of interest $x$ from\npartially observed measurements $y$. In maximum entropy sampling, one selects\nmeasurement locations that are expected to have the highest entropy, so as to\nminimize uncertainty about $x$. This approach relies on an accurate model of\nthe posterior distribution over future measurements, given the measurements\nobserved so far. Recently, diffusion models have been shown to produce\nhigh-quality posterior samples of high-dimensional signals using guided\ndiffusion. In this work, we propose Active Diffusion Subsampling (ADS), a\nmethod for designing intelligent subsampling masks using guided diffusion in\nwhich the model tracks a distribution of beliefs over the true state of $x$\nthroughout the reverse diffusion process, progressively decreasing its\nuncertainty by actively choosing to acquire measurements with maximum expected\nentropy, ultimately producing the posterior distribution $p(x \\mid y)$. ADS can\nbe applied using pre-trained diffusion models for any subsampling rate, and\ndoes not require task-specific retraining - just the specification of a\nmeasurement model. Furthermore, the maximum entropy sampling policy employed by\nADS is interpretable, enhancing transparency relative to existing methods using\nblack-box policies. Code is available at\nhttps://active-diffusion-subsampling.github.io/.\n", "date": "2024-06-20", "categories": []}
{"id": "2504.05490", "title": "Optimal Bayesian Affine Estimator and Active Learning for the Wiener\n  Model", "abstract": "  This paper presents a Bayesian estimation framework for Wiener models,\nfocusing on learning nonlinear output functions under known linear state\ndynamics. We derive a closed-form optimal affine estimator for the unknown\nparameters, characterized by the so-called \"dynamic basis statistics\" (DBS).\nSeveral features of the proposed estimator are studied, including Bayesian\nunbiasedness, closed-form posterior statistics, error monotonicity in\ntrajectory length, and consistency condition (also known as persistent\nexcitation). In the special case of Fourier basis functions, we demonstrate\nthat the closed-form description is computationally available, as the Fourier\nDBS enjoys explicit expressions. Furthermore, we identify an inherent\ninconsistency in the Fourier bases for single-trajectory measurements,\nregardless of the input excitation. Leveraging the closed-form estimation\nerror, we develop an active learning algorithm synthesizing input signals to\nminimize estimation error. Numerical experiments validate the efficacy of our\napproach, showing significant improvements over traditional regularized\nleast-squares methods.\n", "date": "2025-04-07", "categories": []}
{"id": "2411.06353", "title": "Deep Active Learning in the Open World", "abstract": "  Machine learning models deployed in open-world scenarios often encounter\nunfamiliar conditions and perform poorly in unanticipated situations. As AI\nsystems advance and find application in safety-critical domains, effectively\nhandling out-of-distribution (OOD) data is crucial to building open-world\nlearning systems. In this work, we introduce ALOE, a novel active learning\nalgorithm for open-world environments designed to enhance model adaptation by\nincorporating new OOD classes via a two-stage approach. First, diversity\nsampling selects a representative set of examples, followed by energy-based OOD\ndetection to prioritize likely unknown classes for annotation. This strategy\naccelerates class discovery and learning, even under constrained annotation\nbudgets. Evaluations on three long-tailed image classification benchmarks\ndemonstrate that ALOE outperforms traditional active learning baselines,\neffectively expanding known categories while balancing annotation cost. Our\nfindings reveal a crucial tradeoff between enhancing known-class performance\nand discovering new classes, setting the stage for future advancements in\nopen-world machine learning.\n", "date": "2024-11-10", "categories": []}
{"id": "2412.18870", "title": "TSceneJAL: Joint Active Learning of Traffic Scenes for 3D Object\n  Detection", "abstract": "  Most autonomous driving (AD) datasets incur substantial costs for collection\nand labeling, inevitably yielding a plethora of low-quality and redundant data\ninstances, thereby compromising performance and efficiency. Many applications\nin AD systems necessitate high-quality training datasets using both existing\ndatasets and newly collected data. In this paper, we propose a traffic scene\njoint active learning (TSceneJAL) framework that can efficiently sample the\nbalanced, diverse, and complex traffic scenes from both labeled and unlabeled\ndata. The novelty of this framework is threefold: 1) a scene sampling scheme\nbased on a category entropy, to identify scenes containing multiple object\nclasses, thus mitigating class imbalance for the active learner; 2) a\nsimilarity sampling scheme, estimated through the directed graph representation\nand a marginalize kernel algorithm, to pick sparse and diverse scenes; 3) an\nuncertainty sampling scheme, predicted by a mixture density network, to select\ninstances with the most unclear or complex regression outcomes for the learner.\nFinally, the integration of these three schemes in a joint selection strategy\nyields an optimal and valuable subdataset. Experiments on the KITTI, Lyft,\nnuScenes and SUScape datasets demonstrate that our approach outperforms\nexisting state-of-the-art methods on 3D object detection tasks with up to 12%\nimprovements.\n", "date": "2024-12-25", "categories": []}
{"id": "2501.14278", "title": "Active Learning for Continual Learning: Keeping the Past Alive in the\n  Present", "abstract": "  Continual learning (CL) enables deep neural networks to adapt to\never-changing data distributions. In practice, there may be scenarios where\nannotation is costly, leading to active continual learning (ACL), which\nperforms active learning (AL) for the CL scenarios when reducing the labeling\ncost by selecting the most informative subset is preferable. However,\nconventional AL strategies are not suitable for ACL, as they focus solely on\nlearning the new knowledge, leading to catastrophic forgetting of previously\nlearned tasks. Therefore, ACL requires a new AL strategy that can balance the\nprevention of catastrophic forgetting and the ability to quickly learn new\ntasks. In this paper, we propose AccuACL, Accumulated informativeness-based\nActive Continual Learning, by the novel use of the Fisher information matrix as\na criterion for sample selection, derived from a theoretical analysis of the\nFisher-optimality preservation properties within the framework of ACL, while\nalso addressing the scalability issue of Fisher information-based AL. Extensive\nexperiments demonstrate that AccuACL significantly outperforms AL baselines\nacross various CL algorithms, increasing the average accuracy and forgetting by\n23.8% and 17.0%, respectively, on average.\n", "date": "2025-01-24", "categories": []}
{"id": "2503.05981", "title": "Near-Polynomially Competitive Active Logistic Regression", "abstract": "  We address the problem of active logistic regression in the realizable\nsetting. It is well known that active learning can require exponentially fewer\nlabel queries compared to passive learning, in some cases using $\\log\n\\frac{1}{\\eps}$ rather than $\\poly(1/\\eps)$ labels to get error $\\eps$ larger\nthan the optimum.\n  We present the first algorithm that is polynomially competitive with the\noptimal algorithm on every input instance, up to factors polylogarithmic in the\nerror and domain size. In particular, if any algorithm achieves label\ncomplexity polylogarithmic in $\\eps$, so does ours. Our algorithm is based on\nefficient sampling and can be extended to learn more general class of\nfunctions. We further support our theoretical results with experiments\ndemonstrating performance gains for logistic regression compared to existing\nactive learning algorithms.\n", "date": "2025-03-07", "categories": []}
{"id": "2405.10581", "title": "Future Aware Safe Active Learning of Time Varying Systems using Gaussian\n  Processes", "abstract": "  Experimental exploration of high-cost systems with safety constraints, common\nin engineering applications, is a challenging endeavor. Data-driven models\noffer a promising solution, but acquiring the requisite data remains expensive\nand is potentially unsafe. Safe active learning techniques prove essential,\nenabling the learning of high-quality models with minimal expensive data points\nand high safety. This paper introduces a safe active learning framework\ntailored for time-varying systems, addressing drift, seasonal changes, and\ncomplexities due to dynamic behavior. The proposed Time-aware Integrated Mean\nSquared Prediction Error (T-IMSPE) method minimizes posterior variance over\ncurrent and future states, optimizing information gathering also in the time\ndomain. Empirical results highlight T-IMSPE's advantages in model quality\nthrough toy and real-world examples. State of the art Gaussian processes are\ncompatible with T-IMSPE. Our theoretical contributions include a clear\ndelineation which Gaussian process kernels, domains, and weighting measures are\nsuitable for T-IMSPE and even beyond for its non-time aware predecessor IMSPE.\n", "date": "2024-05-17", "categories": []}
{"id": "2409.09475", "title": "MALADY: Multiclass Active Learning with Auction Dynamics on Graphs", "abstract": "  Active learning enhances the performance of machine learning methods,\nparticularly in semi-supervised cases, by judiciously selecting a limited\nnumber of unlabeled data points for labeling, with the goal of improving the\nperformance of an underlying classifier. In this work, we introduce the\nMulticlass Active Learning with Auction Dynamics on Graphs (MALADY) framework\nwhich leverages the auction dynamics algorithm on similarity graphs for\nefficient active learning. In particular, we generalize the auction dynamics\nalgorithm on similarity graphs for semi-supervised learning in [24] to\nincorporate a more general optimization functional. Moreover, we introduce a\nnovel active learning acquisition function that uses the dual variable of the\nauction algorithm to measure the uncertainty in the classifier to prioritize\nqueries near the decision boundaries between different classes. Lastly, using\nexperiments on classification tasks, we evaluate the performance of our\nproposed method and show that it exceeds that of comparison algorithms.\n", "date": "2024-09-14", "categories": []}
{"id": "2504.10559", "title": "Efficient Process Reward Model Training via Active Learning", "abstract": "  Process Reward Models (PRMs) provide step-level supervision to large language\nmodels (LLMs), but scaling up training data annotation remains challenging for\nboth humans and LLMs. To address this limitation, we propose an active learning\napproach, ActPRM, which proactively selects the most uncertain samples for\ntraining, substantially reducing labeling costs. During training, we use the\nPRM to estimate uncertainty after the forward pass, retaining only highly\nuncertain data. A capable yet costly reasoning model then labels this data.\nThen we compute the loss with respect to the labels and update the PRM's\nweights. We compare ActPRM vs. vanilla fine-tuning, on a pool-based active\nlearning setting, demonstrating that ActPRM reduces 50% annotation, but\nachieving the comparable or even better performance. Beyond annotation\nefficiency, we further advance the actively trained PRM by filtering over 1M+\nmath reasoning trajectories with ActPRM, retaining 60% of the data. A\nsubsequent training on this selected dataset yields a new state-of-the-art\n(SOTA) PRM on ProcessBench (75.0%) and PRMBench (65.5%) compared with same\nsized models.\n", "date": "2025-04-14", "categories": []}
{"id": "2408.04917", "title": "Avoid Wasted Annotation Costs in Open-set Active Learning with\n  Pre-trained Vision-Language Model", "abstract": "  Active learning (AL) aims to enhance model performance by selectively\ncollecting highly informative data, thereby minimizing annotation costs.\nHowever, in practical scenarios, unlabeled data may contain out-of-distribution\n(OOD) samples, which are not used for training, leading to wasted annotation\ncosts if data is incorrectly selected. Therefore, to make active learning\nfeasible in real-world applications, it is crucial to consider not only the\ninformativeness of unlabeled samples but also their purity to determine whether\nthey belong to the in-distribution (ID). Recent studies have applied AL under\nthese assumptions, but challenges remain due to the trade-off between\ninformativeness and purity, as well as the heavy dependence on OOD samples.\nThese issues lead to the collection of OOD samples, resulting in a significant\nwaste of annotation costs. To address these challenges, we propose a novel\nquery strategy, VLPure-AL, which minimizes cost losses while reducing\ndependence on OOD samples. VLPure-AL sequentially evaluates the purity and\ninformativeness of data. First, it utilizes a pre-trained vision-language model\nto detect and exclude OOD data with high accuracy by leveraging linguistic and\nvisual information of ID data. Second, it selects highly informative data from\nthe remaining ID data, and then the selected samples are annotated by human\nexperts. Experimental results on datasets with various open-set conditions\ndemonstrate that VLPure-AL achieves the lowest cost loss and highest\nperformance across all scenarios. Code is available at\nhttps://github.com/DSBA-Lab/OpenAL.\n", "date": "2024-08-09", "categories": []}
{"id": "2410.19631", "title": "Efficient Biological Data Acquisition through Inference Set Design", "abstract": "  In drug discovery, highly automated high-throughput laboratories are used to\nscreen a large number of compounds in search of effective drugs. These\nexperiments are expensive, so one might hope to reduce their cost by only\nexperimenting on a subset of the compounds, and predicting the outcomes of the\nremaining experiments. In this work, we model this scenario as a sequential\nsubset selection problem: we aim to select the smallest set of candidates in\norder to achieve some desired level of accuracy for the system as a whole. Our\nkey observation is that, if there is heterogeneity in the difficulty of the\nprediction problem across the input space, selectively obtaining the labels for\nthe hardest examples in the acquisition pool will leave only the relatively\neasy examples to remain in the inference set, leading to better overall system\nperformance. We call this mechanism inference set design, and propose the use\nof a confidence-based active learning solution to prune out these challenging\nexamples. Our algorithm includes an explicit stopping criterion that interrupts\nthe acquisition loop when it is sufficiently confident that the system has\nreached the target performance. Our empirical studies on image and molecular\ndatasets, as well as a real-world large-scale biological assay, show that\nactive learning for inference set design leads to significant reduction in\nexperimental cost while retaining high system performance.\n", "date": "2024-10-25", "categories": []}
{"id": "2504.04506", "title": "Active Learning with a Noisy Annotator", "abstract": "  Active Learning (AL) aims to reduce annotation costs by strategically\nselecting the most informative samples for labeling. However, most active\nlearning methods struggle in the low-budget regime where only a few labeled\nexamples are available. This issue becomes even more pronounced when annotators\nprovide noisy labels. A common AL approach for the low- and mid-budget regimes\nfocuses on maximizing the coverage of the labeled set across the entire\ndataset. We propose a novel framework called Noise-Aware Active Sampling (NAS)\nthat extends existing greedy, coverage-based active learning strategies to\nhandle noisy annotations. NAS identifies regions that remain uncovered due to\nthe selection of noisy representatives and enables resampling from these areas.\nWe introduce a simple yet effective noise filtering approach suitable for the\nlow-budget regime, which leverages the inner mechanism of NAS and can be\napplied for noise filtering before model training. On multiple computer vision\nbenchmarks, including CIFAR100 and ImageNet subsets, NAS significantly improves\nperformance for standard active learning methods across different noise types\nand rates.\n", "date": "2025-04-06", "categories": []}
{"id": "2504.02901", "title": "Hide and Seek in Noise Labels: Noise-Robust Collaborative Active\n  Learning with LLM-Powered Assistance", "abstract": "  Learning from noisy labels (LNL) is a challenge that arises in many\nreal-world scenarios where collected training data can contain incorrect or\ncorrupted labels. Most existing solutions identify noisy labels and adopt\nactive learning to query human experts on them for denoising. In the era of\nlarge language models (LLMs), although we can reduce the human effort to\nimprove these methods, their performances are still subject to accurately\nseparating the clean and noisy samples from noisy data. In this paper, we\npropose an innovative collaborative learning framework NoiseAL based on active\nlearning to combine LLMs and small models (SMs) for learning from noisy labels.\nDuring collaborative training, we first adopt two SMs to form a co-prediction\nnetwork and propose a dynamic-enhanced threshold strategy to divide the noisy\ndata into different subsets, then select the clean and noisy samples from these\nsubsets to feed the active annotator LLMs to rectify noisy samples. Finally, we\nemploy different optimization objectives to conquer subsets with different\ndegrees of label noises. Extensive experiments on synthetic and real-world\nnoise datasets further demonstrate the superiority of our framework over\nstate-of-the-art baselines.\n", "date": "2025-04-03", "categories": []}
{"id": "2503.21095", "title": "Confidence Adjusted Surprise Measure for Active Resourceful Trials\n  (CA-SMART): A Data-driven Active Learning Framework for Accelerating Material\n  Discovery under Resource Constraints", "abstract": "  Accelerating the discovery and manufacturing of advanced materials with\nspecific properties is a critical yet formidable challenge due to vast search\nspace, high costs of experiments, and time-intensive nature of material\ncharacterization. In recent years, active learning, where a surrogate machine\nlearning (ML) model mimics the scientific discovery process of a human\nscientist, has emerged as a promising approach to address these challenges by\nguiding experimentation toward high-value outcomes with a limited budget. Among\nthe diverse active learning philosophies, the concept of surprise (capturing\nthe divergence between expected and observed outcomes) has demonstrated\nsignificant potential to drive experimental trials and refine predictive\nmodels. Scientific discovery often stems from surprise thereby making it a\nnatural driver to guide the search process. Despite its promise, prior studies\nleveraging surprise metrics such as Shannon and Bayesian surprise lack\nmechanisms to account for prior confidence, leading to excessive exploration of\nuncertain regions that may not yield useful information. To address this, we\npropose the Confidence-Adjusted Surprise Measure for Active Resourceful Trials\n(CA-SMART), a novel Bayesian active learning framework tailored for optimizing\ndata-driven experimentation. On a high level, CA-SMART incorporates\nConfidence-Adjusted Surprise (CAS) to dynamically balance exploration and\nexploitation by amplifying surprises in regions where the model is more certain\nwhile discounting them in highly uncertain areas. We evaluated CA-SMART on two\nbenchmark functions (Six-Hump Camelback and Griewank) and in predicting the\nfatigue strength of steel. The results demonstrate superior accuracy and\nefficiency compared to traditional surprise metrics, standard Bayesian\nOptimization (BO) acquisition functions and conventional ML methods.\n", "date": "2025-03-27", "categories": []}
{"id": "2501.15449", "title": "Breaking the SSL-AL Barrier: A Synergistic Semi-Supervised Active\n  Learning Framework for 3D Object Detection", "abstract": "  To address the annotation burden in LiDAR-based 3D object detection, active\nlearning (AL) methods offer a promising solution. However, traditional active\nlearning approaches solely rely on a small amount of labeled data to train an\ninitial model for data selection, overlooking the potential of leveraging the\nabundance of unlabeled data. Recently, attempts to integrate semi-supervised\nlearning (SSL) into AL with the goal of leveraging unlabeled data have faced\nchallenges in effectively resolving the conflict between the two paradigms,\nresulting in less satisfactory performance. To tackle this conflict, we propose\na Synergistic Semi-Supervised Active Learning framework, dubbed as S-SSAL.\nSpecifically, from the perspective of SSL, we propose a Collaborative\nPseudoScene Pre-training (CPSP) method that effectively learns from unlabeled\ndata without introducing adverse effects. From the perspective of AL, we design\na Collaborative Active Learning (CAL) method, which complements the uncertainty\nand diversity methods by model cascading. This allows us to fully exploit the\npotential of the CPSP pre-trained model. Extensive experiments conducted on\nKITTI and Waymo demonstrate the effectiveness of our S-SSAL framework. Notably,\non the KITTI dataset, utilizing only 2% labeled data, S-SSAL can achieve\nperformance comparable to models trained on the full dataset. The code has been\nreleased at https://github.com/LandDreamer/S_SSAL.\n", "date": "2025-01-26", "categories": []}
{"id": "2503.16125", "title": "Uncertainty Meets Diversity: A Comprehensive Active Learning Framework\n  for Indoor 3D Object Detection", "abstract": "  Active learning has emerged as a promising approach to reduce the substantial\nannotation burden in 3D object detection tasks, spurring several initiatives in\noutdoor environments. However, its application in indoor environments remains\nunexplored. Compared to outdoor 3D datasets, indoor datasets face significant\nchallenges, including fewer training samples per class, a greater number of\nclasses, more severe class imbalance, and more diverse scene types and\nintra-class variances. This paper presents the first study on active learning\nfor indoor 3D object detection, where we propose a novel framework tailored for\nthis task. Our method incorporates two key criteria - uncertainty and diversity\n- to actively select the most ambiguous and informative unlabeled samples for\nannotation. The uncertainty criterion accounts for both inaccurate detections\nand undetected objects, ensuring that the most ambiguous samples are\nprioritized. Meanwhile, the diversity criterion is formulated as a joint\noptimization problem that maximizes the diversity of both object class\ndistributions and scene types, using a new Class-aware Adaptive Prototype (CAP)\nbank. The CAP bank dynamically allocates representative prototypes to each\nclass, helping to capture varying intra-class diversity across different\ncategories. We evaluate our method on SUN RGB-D and ScanNetV2, where it\noutperforms baselines by a significant margin, achieving over 85% of\nfully-supervised performance with just 10% of the annotation budget.\n", "date": "2025-03-20", "categories": []}
{"id": "2503.14564", "title": "Effortless Active Labeling for Long-Term Test-Time Adaptation", "abstract": "  Long-term test-time adaptation (TTA) is a challenging task due to error\naccumulation. Recent approaches tackle this issue by actively labeling a small\nproportion of samples in each batch, yet the annotation burden quickly grows as\nthe batch number increases. In this paper, we investigate how to achieve\neffortless active labeling so that a maximum of one sample is selected for\nannotation in each batch. First, we annotate the most valuable sample in each\nbatch based on the single-step optimization perspective in the TTA context. In\nthis scenario, the samples that border between the source- and target-domain\ndata distributions are considered the most feasible for the model to learn in\none iteration. Then, we introduce an efficient strategy to identify these\nsamples using feature perturbation. Second, we discover that the gradient\nmagnitudes produced by the annotated and unannotated samples have significant\nvariations. Therefore, we propose balancing their impact on model optimization\nusing two dynamic weights. Extensive experiments on the popular ImageNet-C, -R,\n-K, -A and PACS databases demonstrate that our approach consistently\noutperforms state-of-the-art methods with significantly lower annotation costs.\n", "date": "2025-03-18", "categories": []}
{"id": "2411.15292", "title": "Influence functions and regularity tangents for efficient active\n  learning", "abstract": "  In this paper we describe an efficient method for providing a regression\nmodel with a sense of curiosity about its data. In the field of machine\nlearning, our framework for representing curiosity is called Active Learning,\nwhich concerns the problem of automatically choosing data points for which to\nquery labels in the semi-supervised setting. The methods we propose are based\non computing a \"regularity tangent\" vector that can be calculated (with only a\nconstant slow-down) together with the model's parameter vector during training.\nWe then take the inner product of this tangent vector with the gradient vector\nof the model's loss at a given data point to obtain a measure of the influence\nof that point on the complexity of the model. In the simplest instantiation,\nthere is only a single regularity tangent vector, of the same dimension as the\nparameter vector. Thus, in the proposed technique, once training is complete,\nevaluating our \"curiosity\" about a potential query data point can be done as\nquickly as calculating the model's loss gradient at that point. The new vector\nonly doubles the amount of storage required by the model. We show that the\nquantity computed by our technique is an example of an \"influence function\",\nand that it measures the expected squared change in model complexity incurred\nby up-weighting a given data point. We propose a number of ways for using this\nand other related quantities to choose new training data points for a\nregression model.\n", "date": "2024-11-22", "categories": []}
{"id": "2412.06470", "title": "Active Learning with Context Sampling and One-vs-Rest Entropy for\n  Semantic Segmentation", "abstract": "  Multi-class semantic segmentation remains a cornerstone challenge in computer\nvision. Yet, dataset creation remains excessively demanding in time and effort,\nespecially for specialized domains. Active Learning (AL) mitigates this\nchallenge by selecting data points for annotation strategically. However,\nexisting patch-based AL methods often overlook boundary pixels critical\ninformation, essential for accurate segmentation. We present OREAL, a novel\npatch-based AL method designed for multi-class semantic segmentation. OREAL\nenhances boundary detection by employing maximum aggregation of pixel-wise\nuncertainty scores. Additionally, we introduce one-vs-rest entropy, a novel\nuncertainty score function that computes class-wise uncertainties while\nachieving implicit class balancing during dataset creation. Comprehensive\nexperiments across diverse datasets and model architectures validate our\nhypothesis.\n", "date": "2024-12-09", "categories": []}
{"id": "2307.08929", "title": "Active learning of effective Hamiltonian for super-large-scale atomic\n  structures", "abstract": "  The first-principles-based effective Hamiltonian scheme provides one of the\nmost accurate modeling technique for large-scale structures, especially for\nferroelectrics. However, the parameterization of the effective Hamiltonian is\ncomplicated and can be difficult for some complex systems such as high-entropy\nperovskites. Here, we propose a general form of effective Hamiltonian and\ndevelop an active machine learning approach to parameterize the effective\nHamiltonian based on Bayesian linear regression. The parameterization is\nemployed in molecular dynamics simulations with the prediction of energy,\nforces, stress and their uncertainties at each step, which decides whether\nfirst-principles calculations are executed to retrain the parameters.\nStructures of BaTiO$_3$, Pb(Zr$_{0.75}$Ti$_{0.25}$)O$_3$ and (Pb,Sr)TiO$_3$\nsystem are taken as examples to show the accuracy of this approach, as compared\nwith conventional parametrization method and experiments. This machine learning\napproach provides a universal and automatic way to compute the effective\nHamiltonian parameters for any considered complex systems with\nsuper-large-scale (more than $10^7$ atoms) atomic structures.\n", "date": "2023-07-18", "categories": []}
{"id": "2502.19691", "title": "Rethinking Epistemic and Aleatoric Uncertainty for Active Open-Set\n  Annotation: An Energy-Based Approach", "abstract": "  Active learning (AL), which iteratively queries the most informative examples\nfrom a large pool of unlabeled candidates for model training, faces significant\nchallenges in the presence of open-set classes. Existing methods either\nprioritize query examples likely to belong to known classes, indicating low\nepistemic uncertainty (EU), or focus on querying those with highly uncertain\npredictions, reflecting high aleatoric uncertainty (AU). However, they both\nyield suboptimal performance, as low EU corresponds to limited useful\ninformation, and closed-set AU metrics for unknown class examples are less\nmeaningful. In this paper, we propose an Energy-based Active Open-set\nAnnotation (EAOA) framework, which effectively integrates EU and AU to achieve\nsuperior performance. EAOA features a $(C+1)$-class detector and a target\nclassifier, incorporating an energy-based EU measure and a margin-based energy\nloss designed for the detector, alongside an energy-based AU measure for the\ntarget classifier. Another crucial component is the target-driven adaptive\nsampling strategy. It first forms a smaller candidate set with low EU scores to\nensure closed-set properties, making AU metrics meaningful. Subsequently,\nexamples with high AU scores are queried to form the final query set, with the\ncandidate set size adjusted adaptively. Extensive experiments show that EAOA\nachieves state-of-the-art performance while maintaining high query precision\nand low training overhead. The code is available at\nhttps://github.com/chenchenzong/EAOA.\n", "date": "2025-02-27", "categories": []}
{"id": "2208.11636", "title": "ImitAL: Learned Active Learning Strategy on Synthetic Data", "abstract": "  Active Learning (AL) is a well-known standard method for efficiently\nobtaining annotated data by first labeling the samples that contain the most\ninformation based on a query strategy. In the past, a large variety of such\nquery strategies has been proposed, with each generation of new strategies\nincreasing the runtime and adding more complexity. However, to the best of our\nour knowledge, none of these strategies excels consistently over a large number\nof datasets from different application domains. Basically, most of the the\nexisting AL strategies are a combination of the two simple heuristics\ninformativeness and representativeness, and the big differences lie in the\ncombination of the often conflicting heuristics. Within this paper, we propose\nImitAL, a domain-independent novel query strategy, which encodes AL as a\nlearning-to-rank problem and learns an optimal combination between both\nheuristics. We train ImitAL on large-scale simulated AL runs on purely\nsynthetic datasets. To show that ImitAL was successfully trained, we perform an\nextensive evaluation comparing our strategy on 13 different datasets, from a\nwide range of domains, with 7 other query strategies.\n", "date": "2022-08-24", "categories": []}
{"id": "2210.03005", "title": "To Softmax, or not to Softmax: that is the question when applying Active\n  Learning for Transformer Models", "abstract": "  Despite achieving state-of-the-art results in nearly all Natural Language\nProcessing applications, fine-tuning Transformer-based language models still\nrequires a significant amount of labeled data to work. A well known technique\nto reduce the amount of human effort in acquiring a labeled dataset is\n\\textit{Active Learning} (AL): an iterative process in which only the minimal\namount of samples is labeled. AL strategies require access to a quantified\nconfidence measure of the model predictions. A common choice is the softmax\nactivation function for the final layer. As the softmax function provides\nmisleading probabilities, this paper compares eight alternatives on seven\ndatasets. Our almost paradoxical finding is that most of the methods are too\ngood at identifying the true most uncertain samples (outliers), and that\nlabeling therefore exclusively outliers results in worse performance. As a\nheuristic we propose to systematically ignore samples, which results in\nimprovements of various methods compared to the softmax function.\n", "date": "2022-10-06", "categories": []}
{"id": "2503.06770", "title": "Unique Rashomon Sets for Robust Active Learning", "abstract": "  Collecting labeled data for machine learning models is often expensive and\ntime-consuming. Active learning addresses this challenge by selectively\nlabeling the most informative observations, but when initial labeled data is\nlimited, it becomes difficult to distinguish genuinely informative points from\nthose appearing uncertain primarily due to noise. Ensemble methods like random\nforests are a powerful approach to quantifying this uncertainty but do so by\naggregating all models indiscriminately. This includes poor performing models\nand redundant models, a problem that worsens in the presence of noisy data. We\nintroduce UNique Rashomon Ensembled Active Learning (UNREAL), which selectively\nensembles only distinct models from the Rashomon set, which is the set of\nnearly optimal models. Restricting ensemble membership to high-performing\nmodels with different explanations helps distinguish genuine uncertainty from\nnoise-induced variation. We show that UNREAL achieves faster theoretical\nconvergence rates than traditional active learning approaches and demonstrates\nempirical improvements of up to 20% in predictive accuracy across five\nbenchmark datasets, while simultaneously enhancing model interpretability.\n", "date": "2025-03-09", "categories": []}
{"id": "2503.06517", "title": "Instance-wise Supervision-level Optimization in Active Learning", "abstract": "  Active learning (AL) is a label-efficient machine learning paradigm that\nfocuses on selectively annotating high-value instances to maximize learning\nefficiency. Its effectiveness can be further enhanced by incorporating weak\nsupervision, which uses rough yet cost-effective annotations instead of exact\n(i.e., full) but expensive annotations. We introduce a novel AL framework,\nInstance-wise Supervision-Level Optimization (ISO), which not only selects the\ninstances to annotate but also determines their optimal annotation level within\na fixed annotation budget. Its optimization criterion leverages the\nvalue-to-cost ratio (VCR) of each instance while ensuring diversity among the\nselected instances. In classification experiments, ISO consistently outperforms\ntraditional AL methods and surpasses a state-of-the-art AL approach that\ncombines full and weak supervision, achieving higher accuracy at a lower\noverall cost. This code is available at\nhttps://github.com/matsuo-shinnosuke/ISOAL.\n", "date": "2025-03-09", "categories": []}
{"id": "2503.07506", "title": "ADROIT: A Self-Supervised Framework for Learning Robust Representations\n  for Active Learning", "abstract": "  Active learning aims to select optimal samples for labeling, minimizing\nannotation costs. This paper introduces a unified representation learning\nframework tailored for active learning with task awareness. It integrates\ndiverse sources, comprising reconstruction, adversarial, self-supervised,\nknowledge-distillation, and classification losses into a unified VAE-based\nADROIT approach. The proposed approach comprises three key components - a\nunified representation generator (VAE), a state discriminator, and a (proxy)\ntask-learner or classifier. ADROIT learns a latent code using both labeled and\nunlabeled data, incorporating task-awareness by leveraging labeled data with\nthe proxy classifier. Unlike previous approaches, the proxy classifier\nadditionally employs a self-supervised loss on unlabeled data and utilizes\nknowledge distillation to align with the target task-learner. The state\ndiscriminator distinguishes between labeled and unlabeled data, facilitating\nthe selection of informative unlabeled samples. The dynamic interaction between\nVAE and the state discriminator creates a competitive environment, with the VAE\nattempting to deceive the discriminator, while the state discriminator learns\nto differentiate between labeled and unlabeled inputs. Extensive evaluations on\ndiverse datasets and ablation analysis affirm the effectiveness of the proposed\nmodel.\n", "date": "2025-03-10", "categories": []}
{"id": "2310.08944", "title": "A Confidence-based Acquisition Model for Self-supervised Active Learning\n  and Label Correction", "abstract": "  Supervised neural approaches are hindered by their dependence on large,\nmeticulously annotated datasets, a requirement that is particularly cumbersome\nfor sequential tasks. The quality of annotations tends to deteriorate with the\ntransition from expert-based to crowd-sourced labelling. To address these\nchallenges, we present CAMEL (Confidence-based Acquisition Model for Efficient\nself-supervised active Learning), a pool-based active learning framework\ntailored to sequential multi-output problems. CAMEL possesses two core\nfeatures: (1) it requires expert annotators to label only a fraction of a\nchosen sequence, and (2) it facilitates self-supervision for the remainder of\nthe sequence. By deploying a label correction mechanism, CAMEL can also be\nutilised for data cleaning. We evaluate CAMEL on two sequential tasks, with a\nspecial emphasis on dialogue belief tracking, a task plagued by the constraints\nof limited and noisy datasets. Our experiments demonstrate that CAMEL\nsignificantly outperforms the baselines in terms of efficiency. Furthermore,\nthe data corrections suggested by our method contribute to an overall\nimprovement in the quality of the resulting datasets.\n", "date": "2023-10-13", "categories": []}
{"id": "2503.02422", "title": "Aggregation Strategies for Efficient Annotation of Bioacoustic Sound\n  Events Using Active Learning", "abstract": "  The vast amounts of audio data collected in Sound Event Detection (SED)\napplications require efficient annotation strategies to enable supervised\nlearning. Manual labeling is expensive and time-consuming, making Active\nLearning (AL) a promising approach for reducing annotation effort. We introduce\nTop K Entropy, a novel uncertainty aggregation strategy for AL that prioritizes\nthe most uncertain segments within an audio recording, instead of averaging\nuncertainty across all segments. This approach enables the selection of entire\nrecordings for annotation, improving efficiency in sparse data scenarios. We\ncompare Top K Entropy to random sampling and Mean Entropy, and show that fewer\nlabels can lead to the same model performance, particularly in datasets with\nsparse sound events. Evaluations are conducted on audio mixtures of sound\nrecordings from parks with meerkat, dog, and baby crying sound events,\nrepresenting real-world bioacoustic monitoring scenarios. Using Top K Entropy\nfor active learning, we can achieve comparable performance to training on the\nfully labeled dataset with only 8% of the labels. Top K Entropy outperforms\nMean Entropy, suggesting that it is best to let the most uncertain segments\nrepresent the uncertainty of an audio file. The findings highlight the\npotential of AL for scalable annotation in audio and time-series applications,\nincluding bioacoustics.\n", "date": "2025-03-04", "categories": []}
{"id": "2503.02491", "title": "Joint Out-of-Distribution Filtering and Data Discovery Active Learning", "abstract": "  As the data demand for deep learning models increases, active learning (AL)\nbecomes essential to strategically select samples for labeling, which maximizes\ndata efficiency and reduces training costs. Real-world scenarios necessitate\nthe consideration of incomplete data knowledge within AL. Prior works address\nhandling out-of-distribution (OOD) data, while another research direction has\nfocused on category discovery. However, a combined analysis of real-world\nconsiderations combining AL with out-of-distribution data and category\ndiscovery remains unexplored. To address this gap, we propose Joint\nOut-of-distribution filtering and data Discovery Active learning (Joda) , to\nuniquely address both challenges simultaneously by filtering out OOD data\nbefore selecting candidates for labeling. In contrast to previous methods, we\ndeeply entangle the training procedure with filter and selection to construct a\ncommon feature space that aligns known and novel categories while separating\nOOD samples. Unlike previous works, Joda is highly efficient and completely\nomits auxiliary models and training access to the unlabeled pool for filtering\nor selection. In extensive experiments on 18 configurations and 3 metrics,\n\\ours{} consistently achieves the highest accuracy with the best class\ndiscovery to OOD filtering balance compared to state-of-the-art competitor\napproaches.\n", "date": "2025-03-04", "categories": []}
{"id": "2503.00867", "title": "DUAL: Diversity and Uncertainty Active Learning for Text Summarization", "abstract": "  With the rise of large language models, neural text summarization has\nadvanced significantly in recent years. However, even state-of-the-art models\ncontinue to rely heavily on high-quality human-annotated data for training and\nevaluation. Active learning is frequently used as an effective way to collect\nsuch datasets, especially when annotation resources are scarce. Active learning\nmethods typically prioritize either uncertainty or diversity but have shown\nlimited effectiveness in summarization, often being outperformed by random\nsampling. We present Diversity and Uncertainty Active Learning (DUAL), a novel\nalgorithm that combines uncertainty and diversity to iteratively select and\nannotate samples that are both representative of the data distribution and\nchallenging for the current model. DUAL addresses the selection of noisy\nsamples in uncertainty-based methods and the limited exploration scope of\ndiversity-based methods. Through extensive experiments with different\nsummarization models and benchmark datasets, we demonstrate that DUAL\nconsistently matches or outperforms the best performing strategies. Using\nvisualizations and quantitative metrics, we provide valuable insights into the\neffectiveness and robustness of different active learning strategies, in an\nattempt to understand why these strategies haven't performed consistently in\ntext summarization. Finally, we show that DUAL strikes a good balance between\ndiversity and robustness.\n", "date": "2025-03-02", "categories": []}
{"id": "2503.01076", "title": "Active Learning for Direct Preference Optimization", "abstract": "  Direct preference optimization (DPO) is a form of reinforcement learning from\nhuman feedback (RLHF) where the policy is learned directly from preferential\nfeedback. Although many models of human preferences exist, the critical task of\nselecting the most informative feedback for training them is under-explored. We\npropose an active learning framework for DPO, which can be applied to collect\nhuman feedback online or to choose the most informative subset of already\ncollected feedback offline. We propose efficient algorithms for both settings.\nThe key idea is to linearize the DPO objective at the last layer of the neural\nnetwork representation of the optimized policy and then compute the D-optimal\ndesign to collect preferential feedback. We prove that the errors in our DPO\nlogit estimates diminish with more feedback. We show the effectiveness of our\nalgorithms empirically in the setting that matches our theory and also on large\nlanguage models.\n", "date": "2025-03-03", "categories": []}
{"id": "2405.01462", "title": "Uncertainty for Active Learning on Graphs", "abstract": "  Uncertainty Sampling is an Active Learning strategy that aims to improve the\ndata efficiency of machine learning models by iteratively acquiring labels of\ndata points with the highest uncertainty. While it has proven effective for\nindependent data its applicability to graphs remains under-explored. We propose\nthe first extensive study of Uncertainty Sampling for node classification: (1)\nWe benchmark Uncertainty Sampling beyond predictive uncertainty and highlight a\nsignificant performance gap to other Active Learning strategies. (2) We develop\nground-truth Bayesian uncertainty estimates in terms of the data generating\nprocess and prove their effectiveness in guiding Uncertainty Sampling toward\noptimal queries. We confirm our results on synthetic data and design an\napproximate approach that consistently outperforms other uncertainty estimators\non real datasets. (3) Based on this analysis, we relate pitfalls in modeling\nuncertainty to existing methods. Our analysis enables and informs the\ndevelopment of principled uncertainty estimation on graphs.\n", "date": "2024-05-02", "categories": []}
{"id": "2412.20644", "title": "Uncertainty Herding: One Active Learning Method for All Label Budgets", "abstract": "  Most active learning research has focused on methods which perform well when\nmany labels are available, but can be dramatically worse than random selection\nwhen label budgets are small. Other methods have focused on the low-budget\nregime, but do poorly as label budgets increase. As the line between \"low\" and\n\"high\" budgets varies by problem, this is a serious issue in practice. We\npropose uncertainty coverage, an objective which generalizes a variety of low-\nand high-budget objectives, as well as natural, hyperparameter-light methods to\nsmoothly interpolate between low- and high-budget regimes. We call greedy\noptimization of the estimate Uncertainty Herding; this simple method is\ncomputationally fast, and we prove that it nearly optimizes the\ndistribution-level coverage. In experimental validation across a variety of\nactive learning tasks, our proposal matches or beats state-of-the-art\nperformance in essentially all cases; it is the only method of which we are\naware that reliably works well in both low- and high-budget settings.\n", "date": "2024-12-30", "categories": []}
{"id": "2409.04328", "title": "Active learning for regression in engineering populations: A\n  risk-informed approach", "abstract": "  Regression is a fundamental prediction task common in data-centric\nengineering applications that involves learning mappings between continuous\nvariables. In many engineering applications (e.g.\\ structural health\nmonitoring), feature-label pairs used to learn such mappings are of limited\navailability which hinders the effectiveness of traditional supervised machine\nlearning approaches. The current paper proposes a methodology for overcoming\nthe issue of data scarcity by combining active learning with hierarchical\nBayesian modelling.\n  Active learning is an approach for preferentially acquiring feature-label\npairs in a resource-efficient manner. In particular, the current work adopts a\nrisk-informed approach that leverages contextual information associated with\nregression-based engineering decision-making tasks (e.g.\\ inspection and\nmaintenance). Hierarchical Bayesian modelling allow multiple related regression\ntasks to be learned over a population, capturing local and global effects. The\ninformation sharing facilitated by this modelling approach means that\ninformation acquired for one engineering system can improve predictive\nperformance across the population.\n  The proposed methodology is demonstrated using an experimental case study.\nSpecifically, multiple regressions are performed over a population of machining\ntools, where the quantity of interest is the surface roughness of the\nworkpieces. An inspection and maintenance decision process is defined using\nthese regression tasks which is in turn used to construct the active-learning\nalgorithm. The novel methodology proposed is benchmarked against an uninformed\napproach to label acquisition and independent modelling of the regression\ntasks. It is shown that the proposed approach has superior performance in terms\nof expected cost -- maintaining predictive performance while reducing the\nnumber of inspections required.\n", "date": "2024-09-06", "categories": []}
{"id": "2502.18142", "title": "Actively Inferring Optimal Measurement Sequences", "abstract": "  Measurement of a physical quantity such as light intensity is an integral\npart of many reconstruction and decision scenarios but can be costly in terms\nof acquisition time, invasion of or damage to the environment and storage. Data\nminimisation and compliance with data protection laws is also an important\nconsideration. Where there are a range of measurements that can be made, some\nmay be more informative and compliant with the overall measurement objective\nthan others. We develop an active sequential inference algorithm that uses the\nlow dimensional representational latent space from a variational autoencoder\n(VAE) to choose which measurement to make next. Our aim is to recover high\ndimensional data by making as few measurements as possible. We adapt the VAE\nencoder to map partial data measurements on to the latent space of the complete\ndata. The algorithm draws samples from this latent space and uses the VAE\ndecoder to generate data conditional on the partial measurements. Estimated\nmeasurements are made on the generated data and fed back through the partial\nVAE encoder to the latent space where they can be evaluated prior to making a\nmeasurement. Starting from no measurements and a normal prior on the latent\nspace, we consider alternative strategies for choosing the next measurement and\nupdating the predictive posterior prior for the next step. The algorithm is\nillustrated using the Fashion MNIST dataset and a novel convolutional Hadamard\npattern measurement basis. We see that useful patterns are chosen within 10\nsteps, leading to the convergence of the guiding generative images. Compared\nwith using stochastic variational inference to infer the parameters of the\nposterior distribution for each generated data point individually, the partial\nVAE framework can efficiently process batches of generated data and obtains\nsuperior results with minimal measurements.\n", "date": "2025-02-25", "categories": []}
{"id": "2502.18213", "title": "Near-optimal Active Regression of Single-Index Models", "abstract": "  The active regression problem of the single-index model is to solve $\\min_x\n\\lVert f(Ax)-b\\rVert_p$, where $A$ is fully accessible and $b$ can only be\naccessed via entry queries, with the goal of minimizing the number of queries\nto the entries of $b$. When $f$ is Lipschitz, previous results only obtain\nconstant-factor approximations. This work presents the first algorithm that\nprovides a $(1+\\varepsilon)$-approximation solution by querying\n$\\tilde{O}(d^{\\frac{p}{2}\\vee 1}/\\varepsilon^{p\\vee 2})$ entries of $b$. This\nquery complexity is also shown to be optimal up to logarithmic factors for\n$p\\in [1,2]$ and the $\\varepsilon$-dependence of $1/\\varepsilon^p$ is shown to\nbe optimal for $p>2$.\n", "date": "2025-02-25", "categories": []}
{"id": "2406.04273", "title": "ELFS: Label-Free Coreset Selection with Proxy Training Dynamics", "abstract": "  High-quality human-annotated data is crucial for modern deep learning\npipelines, yet the human annotation process is both costly and time-consuming.\nGiven a constrained human labeling budget, selecting an informative and\nrepresentative data subset for labeling can significantly reduce human\nannotation effort. Well-performing state-of-the-art (SOTA) coreset selection\nmethods require ground truth labels over the whole dataset, failing to reduce\nthe human labeling burden. Meanwhile, SOTA label-free coreset selection methods\ndeliver inferior performance due to poor geometry-based difficulty scores. In\nthis paper, we introduce ELFS (Effective Label-Free Coreset Selection), a novel\nlabel-free coreset selection method. ELFS significantly improves label-free\ncoreset selection by addressing two challenges: 1) ELFS utilizes deep\nclustering to estimate training dynamics-based data difficulty scores without\nground truth labels; 2) Pseudo-labels introduce a distribution shift in the\ndata difficulty scores, and we propose a simple but effective double-end\npruning method to mitigate bias on calculated scores. We evaluate ELFS on four\nvision benchmarks and show that, given the same vision encoder, ELFS\nconsistently outperforms SOTA label-free baselines. For instance, when using\nSwAV as the encoder, ELFS outperforms D2 by up to 10.2% in accuracy on\nImageNet-1K. We make our code publicly available on GitHub.\n", "date": "2024-06-06", "categories": []}
{"id": "2502.16425", "title": "Active Learning Classification from a Signal Separation Perspective", "abstract": "  In machine learning, classification is usually seen as a function\napproximation problem, where the goal is to learn a function that maps input\nfeatures to class labels. In this paper, we propose a novel clustering and\nclassification framework inspired by the principles of signal separation. This\napproach enables efficient identification of class supports, even in the\npresence of overlapping distributions. We validate our method on real-world\nhyperspectral datasets Salinas and Indian Pines. The experimental results\ndemonstrate that our method is competitive with the state of the art active\nlearning algorithms by using a very small subset of data set as training\npoints.\n", "date": "2025-02-23", "categories": []}
{"id": "2502.16984", "title": "Active Learning for Conditional Inverse Design with Crystal Generation\n  and Foundation Atomic Models", "abstract": "  Artificial intelligence (AI) is transforming materials science, enabling both\ntheoretical advancements and accelerated materials discovery. Recent progress\nin crystal generation models, which design crystal structures for targeted\nproperties, and foundation atomic models (FAMs), which capture interatomic\ninteractions across the periodic table, has significantly improved inverse\nmaterials design. However, an efficient integration of these two approaches\nremains an open challenge. Here, we present an active learning framework that\ncombines crystal generation models and foundation atomic models to enhance the\naccuracy and efficiency of inverse design. As a case study, we employ Con-CDVAE\nto generate candidate crystal structures and MACE-MP-0 FAM as one of the\nhigh-throughput screeners for bulk modulus evaluation. Through iterative active\nlearning, we demonstrate that Con-CDVAE progressively improves its accuracy in\ngenerating crystals with target properties, highlighting the effectiveness of a\nproperty-driven fine-tuning process. Our framework is general to accommodate\ndifferent crystal generation and foundation atomic models, and establishes a\nscalable approach for AI-driven materials discovery. By bridging generative\nmodeling with atomic-scale simulations, this work paves the way for more\naccurate and efficient inverse materials design.\n", "date": "2025-02-24", "categories": []}
{"id": "2311.18356", "title": "Towards Comparable Active Learning", "abstract": "  Active Learning has received significant attention in the field of machine\nlearning for its potential in selecting the most informative samples for\nlabeling, thereby reducing data annotation costs. However, we show that the\nreported lifts in recent literature generalize poorly to other domains leading\nto an inconclusive landscape in Active Learning research. Furthermore, we\nhighlight overlooked problems for reproducing AL experiments that can lead to\nunfair comparisons and increased variance in the results. This paper addresses\nthese issues by providing an Active Learning framework for a fair comparison of\nalgorithms across different tasks and domains, as well as a fast and performant\noracle algorithm for evaluation. To the best of our knowledge, we propose the\nfirst AL benchmark that tests algorithms in 3 major domains: Tabular, Image,\nand Text. We report empirical results for 6 widely used algorithms on 7\nreal-world and 2 synthetic datasets and aggregate them into a domain-specific\nranking of AL algorithms.\n", "date": "2023-11-30", "categories": []}
{"id": "2502.15643", "title": "AutoTandemML: Active Learning Enhanced Tandem Neural Networks for\n  Inverse Design Problems", "abstract": "  Inverse design in science and engineering involves determining optimal design\nparameters that achieve desired performance outcomes, a process often hindered\nby the complexity and high dimensionality of design spaces, leading to\nsignificant computational costs. To tackle this challenge, we propose a novel\nhybrid approach that combines active learning with Tandem Neural Networks to\nenhance the efficiency and effectiveness of solving inverse design problems.\nActive learning allows to selectively sample the most informative data points,\nreducing the required dataset size without compromising accuracy. We\ninvestigate this approach using three benchmark problems: airfoil inverse\ndesign, photonic surface inverse design, and scalar boundary condition\nreconstruction in diffusion partial differential equations. We demonstrate that\nintegrating active learning with Tandem Neural Networks outperforms standard\napproaches across the benchmark suite, achieving better accuracy with fewer\ntraining samples.\n", "date": "2025-02-21", "categories": []}
{"id": "2502.14842", "title": "Generating $\\pi$-Functional Molecules Using STGG+ with Active Learning", "abstract": "  Generating novel molecules with out-of-distribution properties is a major\nchallenge in molecular discovery. While supervised learning methods generate\nhigh-quality molecules similar to those in a dataset, they struggle to\ngeneralize to out-of-distribution properties. Reinforcement learning can\nexplore new chemical spaces but often conducts 'reward-hacking' and generates\nnon-synthesizable molecules. In this work, we address this problem by\nintegrating a state-of-the-art supervised learning method, STGG+, in an active\nlearning loop. Our approach iteratively generates, evaluates, and fine-tunes\nSTGG+ to continuously expand its knowledge. We denote this approach STGG+AL. We\napply STGG+AL to the design of organic $\\pi$-functional materials, specifically\ntwo challenging tasks: 1) generating highly absorptive molecules characterized\nby high oscillator strength and 2) designing absorptive molecules with\nreasonable oscillator strength in the near-infrared (NIR) range. The generated\nmolecules are validated and rationalized in-silico with time-dependent density\nfunctional theory. Our results demonstrate that our method is highly effective\nin generating novel molecules with high oscillator strength, contrary to\nexisting methods such as reinforcement learning (RL) methods. We open-source\nour active-learning code along with our Conjugated-xTB dataset containing 2.9\nmillion $\\pi$-conjugated molecules and the function for approximating the\noscillator strength and absorption wavelength (based on sTDA-xTB).\n", "date": "2025-02-20", "categories": []}
{"id": "2207.06030", "title": "Contextual Active Model Selection", "abstract": "  While training models and labeling data are resource-intensive, a wealth of\npre-trained models and unlabeled data exists. To effectively utilize these\nresources, we present an approach to actively select pre-trained models while\nminimizing labeling costs. We frame this as an online contextual active model\nselection problem: At each round, the learner receives an unlabeled data point\nas a context. The objective is to adaptively select the best model to make a\nprediction while limiting label requests. To tackle this problem, we propose\nCAMS, a contextual active model selection algorithm that relies on two novel\ncomponents: (1) a contextual model selection mechanism, which leverages context\ninformation to make informed decisions about which model is likely to perform\nbest for a given context, and (2) an active query component, which\nstrategically chooses when to request labels for data points, minimizing the\noverall labeling cost. We provide rigorous theoretical analysis for the regret\nand query complexity under both adversarial and stochastic settings.\nFurthermore, we demonstrate the effectiveness of our algorithm on a diverse\ncollection of benchmark classification tasks. Notably, CAMS requires\nsubstantially less labeling effort (less than 10%) compared to existing methods\non CIFAR10 and DRIFT benchmarks, while achieving similar or better accuracy.\nOur code is publicly available at:\nhttps://github.com/xuefeng-cs/Contextual-Active-Model-Selection.\n", "date": "2022-07-13", "categories": []}
{"id": "2402.15898", "title": "Transductive Active Learning: Theory and Applications", "abstract": "  We study a generalization of classical active learning to real-world settings\nwith concrete prediction targets where sampling is restricted to an accessible\nregion of the domain, while prediction targets may lie outside this region. We\nanalyze a family of decision rules that sample adaptively to minimize\nuncertainty about prediction targets. We are the first to show, under general\nregularity assumptions, that such decision rules converge uniformly to the\nsmallest possible uncertainty obtainable from the accessible data. We\ndemonstrate their strong sample efficiency in two key applications: active\nfine-tuning of large neural networks and safe Bayesian optimization, where they\nachieve state-of-the-art performance.\n", "date": "2024-02-13", "categories": []}
{"id": "2502.06076", "title": "A Planning Framework for Adaptive Labeling", "abstract": "  Ground truth labels/outcomes are critical for advancing scientific and\nengineering applications, e.g., evaluating the treatment effect of an\nintervention or performance of a predictive model. Since randomly sampling\ninputs for labeling can be prohibitively expensive, we introduce an adaptive\nlabeling framework where measurement effort can be reallocated in batches. We\nformulate this problem as a Markov decision process where posterior beliefs\nevolve over time as batches of labels are collected (state transition), and\nbatches (actions) are chosen to minimize uncertainty at the end of data\ncollection. We design a computational framework that is agnostic to different\nuncertainty quantification approaches including those based on deep learning,\nand allows a diverse array of policy gradient approaches by relying on\ncontinuous policy parameterizations. On real and synthetic datasets, we\ndemonstrate even a one-step lookahead policy can substantially outperform\ncommon adaptive labeling heuristics, highlighting the virtue of planning. On\nthe methodological side, we note that standard REINFORCE-style policy gradient\nestimators can suffer high variance since they rely only on zeroth order\ninformation. We propose a direct backpropagation-based approach,\nSmoothed-Autodiff, based on a carefully smoothed version of the original\nnon-differentiable MDP. Our method enjoys low variance at the price of\nintroducing bias, and we theoretically and empirically show that this trade-off\ncan be favorable.\n", "date": "2025-02-10", "categories": []}
{"id": "2502.03292", "title": "ALPET: Active Few-shot Learning for Citation Worthiness Detection in\n  Low-Resource Wikipedia Languages", "abstract": "  Citation Worthiness Detection (CWD) consists in determining which sentences,\nwithin an article or collection, should be backed up with a citation to\nvalidate the information it provides. This study, introduces ALPET, a framework\ncombining Active Learning (AL) and Pattern-Exploiting Training (PET), to\nenhance CWD for languages with limited data resources. Applied to Catalan,\nBasque, and Albanian Wikipedia datasets, ALPET outperforms the existing CCW\nbaseline while reducing the amount of labeled data in some cases above 80\\%.\nALPET's performance plateaus after 300 labeled samples, showing it suitability\nfor low-resource scenarios where large, labeled datasets are not common. While\nspecific active learning query strategies, like those employing K-Means\nclustering, can offer advantages, their effectiveness is not universal and\noften yields marginal gains over random sampling, particularly with smaller\ndatasets. This suggests that random sampling, despite its simplicity, remains a\nstrong baseline for CWD in constraint resource environments. Overall, ALPET's\nability to achieve high performance with fewer labeled samples makes it a\npromising tool for enhancing the verifiability of online content in\nlow-resource language settings.\n", "date": "2025-02-05", "categories": []}
{"id": "2409.01416", "title": "Active Symbolic Discovery of Ordinary Differential Equations via Phase\n  Portrait Sketching", "abstract": "  The symbolic discovery of Ordinary Differential Equations (ODEs) from\ntrajectory data plays a pivotal role in AI-driven scientific discovery.\nExisting symbolic methods predominantly rely on fixed, pre-collected training\ndatasets, which often result in suboptimal performance, as demonstrated in our\ncase study in Figure 1. Drawing inspiration from active learning, we\ninvestigate strategies to query informative trajectory data that can enhance\nthe evaluation of predicted ODEs. However, the butterfly effect in dynamical\nsystems reveals that small variations in initial conditions can lead to\ndrastically different trajectories, necessitating the storage of vast\nquantities of trajectory data using conventional active learning. To address\nthis, we introduce Active Symbolic Discovery of Ordinary Differential Equations\nvia Phase Portrait Sketching (APPS). Instead of directly selecting individual\ninitial conditions, our APPS first identifies an informative region within the\nphase space and then samples a batch of initial conditions from this region.\nCompared to traditional active learning methods, APPS mitigates the gap of\nmaintaining a large amount of data. Extensive experiments demonstrate that APPS\nconsistently discovers more accurate ODE expressions than baseline methods\nusing passively collected datasets.\n", "date": "2024-09-02", "categories": []}
{"id": "2502.00305", "title": "DEUCE: Dual-diversity Enhancement and Uncertainty-awareness for\n  Cold-start Active Learning", "abstract": "  Cold-start active learning (CSAL) selects valuable instances from an\nunlabeled dataset for manual annotation. It provides high-quality data at a low\nannotation cost for label-scarce text classification. However, existing CSAL\nmethods overlook weak classes and hard representative examples, resulting in\nbiased learning. To address these issues, this paper proposes a novel\ndual-diversity enhancing and uncertainty-aware (DEUCE) framework for CSAL.\nSpecifically, DEUCE leverages a pretrained language model (PLM) to efficiently\nextract textual representations, class predictions, and predictive uncertainty.\nThen, it constructs a Dual-Neighbor Graph (DNG) to combine information on both\ntextual diversity and class diversity, ensuring a balanced data distribution.\nIt further propagates uncertainty information via density-based clustering to\nselect hard representative instances. DEUCE performs well in selecting\nclass-balanced and hard representative data by dual-diversity and\ninformativeness. Experiments on six NLP datasets demonstrate the superiority\nand efficiency of DEUCE.\n", "date": "2025-02-01", "categories": []}
{"id": "2502.01012", "title": "Deep Active Learning based Experimental Design to Uncover Synergistic\n  Genetic Interactions for Host Targeted Therapeutics", "abstract": "  Recent technological advances have introduced new high-throughput methods for\nstudying host-virus interactions, but testing synergistic interactions between\nhost gene pairs during infection remains relatively slow and labor intensive.\nIdentification of multiple gene knockdowns that effectively inhibit viral\nreplication requires a search over the combinatorial space of all possible\ntarget gene pairs and is infeasible via brute-force experiments. Although\nactive learning methods for sequential experimental design have shown promise,\nexisting approaches have generally been restricted to single-gene knockdowns or\nsmall-scale double knockdown datasets. In this study, we present an integrated\nDeep Active Learning (DeepAL) framework that incorporates information from a\nbiological knowledge graph (SPOKE, the Scalable Precision Medicine Open\nKnowledge Engine) to efficiently search the configuration space of a large\ndataset of all pairwise knockdowns of 356 human genes in HIV infection. Through\ngraph representation learning, the framework is able to generate task-specific\nrepresentations of genes while also balancing the exploration-exploitation\ntrade-off to pinpoint highly effective double-knockdown pairs. We additionally\npresent an ensemble method for uncertainty quantification and an interpretation\nof the gene pairs selected by our algorithm via pathway analysis. To our\nknowledge, this is the first work to show promising results on double-gene\nknockdown experimental data of appreciable scale (356 by 356 matrix).\n", "date": "2025-02-03", "categories": []}
{"id": "2410.00542", "title": "Differentially Private Active Learning: Balancing Effective Data\n  Selection and Privacy", "abstract": "  Active learning (AL) is a widely used technique for optimizing data labeling\nin machine learning by iteratively selecting, labeling, and training on the\nmost informative data. However, its integration with formal privacy-preserving\nmethods, particularly differential privacy (DP), remains largely underexplored.\nWhile some works have explored differentially private AL for specialized\nscenarios like online learning, the fundamental challenge of combining AL with\nDP in standard learning settings has remained unaddressed, severely limiting\nAL's applicability in privacy-sensitive domains. This work addresses this gap\nby introducing differentially private active learning (DP-AL) for standard\nlearning settings. We demonstrate that naively integrating DP-SGD training into\nAL presents substantial challenges in privacy budget allocation and data\nutilization. To overcome these challenges, we propose step amplification, which\nleverages individual sampling probabilities in batch creation to maximize data\npoint participation in training steps, thus optimizing data utilization.\nAdditionally, we investigate the effectiveness of various acquisition functions\nfor data selection under privacy constraints, revealing that many commonly used\nfunctions become impractical. Our experiments on vision and natural language\nprocessing tasks show that DP-AL can improve performance for specific datasets\nand model architectures. However, our findings also highlight the limitations\nof AL in privacy-constrained environments, emphasizing the trade-offs between\nprivacy, model accuracy, and data selection accuracy.\n", "date": "2024-10-01", "categories": []}
{"id": "2305.06584", "title": "Active Learning For Contextual Linear Optimization: A Margin-Based\n  Approach", "abstract": "  We develop the first active learning method for contextual linear\noptimization. Specifically, we introduce a label acquisition algorithm that\nsequentially decides whether to request the ``labels'' of feature samples from\nan unlabeled data stream, where the labels correspond to the coefficients of\nthe objective in the linear optimization. Our method is the first to be\ndirectly informed by the decision loss induced by the predicted coefficients,\nreferred to as the Smart Predict-then-Optimize (SPO) loss. Motivated by the\nstructure of the SPO loss, our algorithm adopts a margin-based criterion\nutilizing the concept of distance to degeneracy. In particular, we design an\nefficient active learning algorithm with theoretical excess risk (i.e.,\ngeneralization) guarantees. We derive upper bounds on the label complexity,\ndefined as the number of samples whose labels are acquired to achieve a desired\nsmall level of SPO risk. These bounds show that our algorithm has a much\nsmaller label complexity than the naive supervised learning approach that\nlabels all samples, particularly when the SPO loss is minimized directly on the\ncollected data. To address the discontinuity and nonconvexity of the SPO loss,\nwe derive label complexity bounds under tractable surrogate loss functions.\nUnder natural margin conditions, these bounds also outperform naive supervised\nlearning. Using the SPO+ loss, a specialized surrogate of the SPO loss, we\nestablish even tighter bounds under separability conditions. Finally, we\npresent numerical evidence showing the practical value of our algorithms in\nsettings such as personalized pricing and the shortest path problem.\n", "date": "2023-05-11", "categories": []}
{"id": "2501.13517", "title": "Propensity-driven Uncertainty Learning for Sample Exploration in\n  Source-Free Active Domain Adaptation", "abstract": "  Source-free active domain adaptation (SFADA) addresses the challenge of\nadapting a pre-trained model to new domains without access to source data while\nminimizing the need for target domain annotations. This scenario is\nparticularly relevant in real-world applications where data privacy, storage\nlimitations, or labeling costs are significant concerns. Key challenges in\nSFADA include selecting the most informative samples from the target domain for\nlabeling, effectively leveraging both labeled and unlabeled target data, and\nadapting the model without relying on source domain information. Additionally,\nexisting methods often struggle with noisy or outlier samples and may require\nimpractical progressive labeling during training. To effectively select more\ninformative samples without frequently requesting human annotations, we propose\nthe Propensity-driven Uncertainty Learning (ProULearn) framework. ProULearn\nutilizes a novel homogeneity propensity estimation mechanism combined with\ncorrelation index calculation to evaluate feature-level relationships. This\napproach enables the identification of representative and challenging samples\nwhile avoiding noisy outliers. Additionally, we develop a central correlation\nloss to refine pseudo-labels and create compact class distributions during\nadaptation. In this way, ProULearn effectively bridges the domain gap and\nmaximizes adaptation performance. The principles of informative sample\nselection underlying ProULearn have broad implications beyond SFADA, offering\nbenefits across various deep learning tasks where identifying key data points\nor features is crucial. Extensive experiments on four benchmark datasets\ndemonstrate that ProULearn outperforms state-of-the-art methods in domain\nadaptation scenarios.\n", "date": "2025-01-23", "categories": []}
{"id": "2403.03728", "title": "Bridging Diversity and Uncertainty in Active learning with\n  Self-Supervised Pre-Training", "abstract": "  This study addresses the integration of diversity-based and uncertainty-based\nsampling strategies in active learning, particularly within the context of\nself-supervised pre-trained models. We introduce a straightforward heuristic\ncalled TCM that mitigates the cold start problem while maintaining strong\nperformance across various data levels. By initially applying TypiClust for\ndiversity sampling and subsequently transitioning to uncertainty sampling with\nMargin, our approach effectively combines the strengths of both strategies. Our\nexperiments demonstrate that TCM consistently outperforms existing methods\nacross various datasets in both low and high data regimes.\n", "date": "2024-03-06", "categories": []}
{"id": "2501.08450", "title": "Active Sampling for Node Attribute Completion on Graphs", "abstract": "  Node attribute, a type of crucial information for graph analysis, may be\npartially or completely missing for certain nodes in real world applications.\nRestoring the missing attributes is expected to benefit downstream graph\nlearning. Few attempts have been made on node attribute completion, but a novel\nframework called Structure-attribute Transformer (SAT) was recently proposed by\nusing a decoupled scheme to leverage structures and attributes. SAT ignores the\ndifferences in contributing to the learning schedule and finding a practical\nway to model the different importance of nodes with observed attributes is\nchallenging. This paper proposes a novel AcTive Sampling algorithm (ATS) to\nrestore missing node attributes. The representativeness and uncertainty of each\nnode's information are first measured based on graph structure, representation\nsimilarity and learning bias. To select nodes as train samples in the next\noptimization step, a weighting scheme controlled by Beta distribution is then\nintroduced to linearly combine the two properties. Extensive experiments on\nfour public benchmark datasets and two downstream tasks have shown the\nsuperiority of ATS in node attribute completion.\n", "date": "2025-01-14", "categories": []}
{"id": "2501.08223", "title": "Big Batch Bayesian Active Learning by Considering Predictive\n  Probabilities", "abstract": "  We observe that BatchBALD, a popular acquisition function for batch Bayesian\nactive learning for classification, can conflate epistemic and aleatoric\nuncertainty, leading to suboptimal performance. Motivated by this observation,\nwe propose to focus on the predictive probabilities, which only exhibit\nepistemic uncertainty. The result is an acquisition function that not only\nperforms better, but is also faster to evaluate, allowing for larger batches\nthan before.\n", "date": "2025-01-14", "categories": []}
{"id": "2501.08245", "title": "Continual Deep Active Learning for Medical Imaging: Replay-Base\n  Architecture for Context Adaptation", "abstract": "  Deep Learning for medical imaging faces challenges in adapting and\ngeneralizing to new contexts. Additionally, it often lacks sufficient labeled\ndata for specific tasks requiring significant annotation effort. Continual\nLearning (CL) tackles adaptability and generalizability by enabling lifelong\nlearning from a data stream while mitigating forgetting of previously learned\nknowledge. Active Learning (AL) reduces the number of required annotations for\neffective training. This work explores both approaches (CAL) to develop a novel\nframework for robust medical image analysis. Based on the automatic recognition\nof shifts in image characteristics, Replay-Base Architecture for Context\nAdaptation (RBACA) employs a CL rehearsal method to continually learn from\ndiverse contexts, and an AL component to select the most informative instances\nfor annotation. A novel approach to evaluate CAL methods is established using a\ndefined metric denominated IL-Score, which allows for the simultaneous\nassessment of transfer learning, forgetting, and final model performance. We\nshow that RBACA works in domain and class-incremental learning scenarios, by\nassessing its IL-Score on the segmentation and diagnosis of cardiac images. The\nresults show that RBACA outperforms a baseline framework without CAL, and a\nstate-of-the-art CAL method across various memory sizes and annotation budgets.\nOur code is available in https://github.com/RuiDaniel/RBACA .\n", "date": "2025-01-14", "categories": []}
{"id": "2407.17992", "title": "Amortized Active Learning for Nonparametric Functions", "abstract": "  Active learning (AL) is a sequential learning scheme aiming to select the\nmost informative data. AL reduces data consumption and avoids the cost of\nlabeling large amounts of data. However, AL trains the model and solves an\nacquisition optimization for each selection. It becomes expensive when the\nmodel training or acquisition optimization is challenging. In this paper, we\nfocus on active nonparametric function learning, where the gold standard\nGaussian process (GP) approaches suffer from cubic time complexity. We propose\nan amortized AL method, where new data are suggested by a neural network which\nis trained up-front without any real data (Figure 1). Our method avoids\nrepeated model training and requires no acquisition optimization during the AL\ndeployment. We (i) utilize GPs as function priors to construct an AL simulator,\n(ii) train an AL policy that can zero-shot generalize from simulation to real\nlearning problems of nonparametric functions and (iii) achieve real-time data\nselection and comparable learning performances to time-consuming baseline\nmethods.\n", "date": "2024-07-25", "categories": []}
{"id": "2501.00149", "title": "LASSE: Learning Active Sampling for Storm Tide Extremes in\n  Non-Stationary Climate Regimes", "abstract": "  Identifying tropical cyclones that generate destructive storm tides for risk\nassessment, such as from large downscaled storm catalogs for climate studies,\nis often intractable because it entails many expensive Monte Carlo hydrodynamic\nsimulations. Here, we show that surrogate models are promising from accuracy,\nrecall, and precision perspectives, and they \"generalize\" to novel climate\nscenarios. We then present an informative online learning approach to rapidly\nsearch for extreme storm tide-producing cyclones using only a few hydrodynamic\nsimulations. Starting from a minimal subset of TCs with detailed storm tide\nhydrodynamic simulations, a surrogate model selects informative data to retrain\nonline and iteratively improves its predictions of damaging TCs. Results on an\nextensive catalog of downscaled TCs indicate 100% precision in retrieving rare\ndestructive storms using less than 20% of the simulations as training. The\ninformative sampling approach is efficient, scalable to large storm catalogs,\nand generalizable to climate scenarios.\n", "date": "2024-12-30", "categories": []}
{"id": "2501.00508", "title": "Active Learning of General Halfspaces: Label Queries vs Membership\n  Queries", "abstract": "  We study the problem of learning general (i.e., not necessarily homogeneous)\nhalfspaces under the Gaussian distribution on $R^d$ in the presence of some\nform of query access. In the classical pool-based active learning model, where\nthe algorithm is allowed to make adaptive label queries to previously sampled\npoints, we establish a strong information-theoretic lower bound ruling out\nnon-trivial improvements over the passive setting. Specifically, we show that\nany active learner requires label complexity of\n$\\tilde{\\Omega}(d/(\\log(m)\\epsilon))$, where $m$ is the number of unlabeled\nexamples. Specifically, to beat the passive label complexity of $\\tilde{O}\n(d/\\epsilon)$, an active learner requires a pool of $2^{poly(d)}$ unlabeled\nsamples. On the positive side, we show that this lower bound can be\ncircumvented with membership query access, even in the agnostic model.\nSpecifically, we give a computationally efficient learner with query complexity\nof $\\tilde{O}(\\min\\{1/p, 1/\\epsilon\\} + d\\cdot polylog(1/\\epsilon))$ achieving\nerror guarantee of $O(opt)+\\epsilon$. Here $p \\in [0, 1/2]$ is the bias and\n$opt$ is the 0-1 loss of the optimal halfspace. As a corollary, we obtain a\nstrong separation between the active and membership query models. Taken\ntogether, our results characterize the complexity of learning general\nhalfspaces under Gaussian marginals in these models.\n", "date": "2024-12-31", "categories": []}
{"id": "2501.01248", "title": "Bayesian Active Learning By Distribution Disagreement", "abstract": "  Active Learning (AL) for regression has been systematically under-researched\ndue to the increased difficulty of measuring uncertainty in regression models.\nSince normalizing flows offer a full predictive distribution instead of a point\nforecast, they facilitate direct usage of known heuristics for AL like Entropy\nor Least-Confident sampling. However, we show that most of these heuristics do\nnot work well for normalizing flows in pool-based AL and we need more\nsophisticated algorithms to distinguish between aleatoric and epistemic\nuncertainty. In this work we propose BALSA, an adaptation of the BALD\nalgorithm, tailored for regression with normalizing flows. With this work we\nextend current research on uncertainty quantification with normalizing flows\n\\cite{berry2023normalizing, berry2023escaping} to real world data and\npool-based AL with multiple acquisition functions and query sizes. We report\nSOTA results for BALSA across 4 different datasets and 2 different\narchitectures.\n", "date": "2025-01-02", "categories": []}
{"id": "2412.19877", "title": "Image Classification with Deep Reinforcement Active Learning", "abstract": "  Deep learning is currently reaching outstanding performances on different\ntasks, including image classification, especially when using large neural\nnetworks. The success of these models is tributary to the availability of large\ncollections of labeled training data. In many real-world scenarios, labeled\ndata are scarce, and their hand-labeling is time, effort and cost demanding.\nActive learning is an alternative paradigm that mitigates the effort in\nhand-labeling data, where only a small fraction is iteratively selected from a\nlarge pool of unlabeled data, and annotated by an expert (a.k.a oracle), and\neventually used to update the learning models. However, existing active\nlearning solutions are dependent on handcrafted strategies that may fail in\nhighly variable learning environments (datasets, scenarios, etc). In this work,\nwe devise an adaptive active learning method based on Markov Decision Process\n(MDP). Our framework leverages deep reinforcement learning and active learning\ntogether with a Deep Deterministic Policy Gradient (DDPG) in order to\ndynamically adapt sample selection strategies to the oracle's feedback and the\nlearning environment. Extensive experiments conducted on three different image\nclassification benchmarks show superior performances against several existing\nactive learning strategies.\n", "date": "2024-12-27", "categories": []}
{"id": "2404.04736", "title": "ProtoAL: Interpretable Deep Active Learning with prototypes for medical\n  imaging", "abstract": "  The adoption of Deep Learning algorithms in the medical imaging field is a\nprominent area of research, with high potential for advancing AI-based\nComputer-aided diagnosis (AI-CAD) solutions. However, current solutions face\nchallenges due to a lack of interpretability features and high data demands,\nprompting recent efforts to address these issues. In this study, we propose the\nProtoAL method, where we integrate an interpretable DL model into the Deep\nActive Learning (DAL) framework. This approach aims to address both challenges\nby focusing on the medical imaging context and utilizing an inherently\ninterpretable model based on prototypes. We evaluated ProtoAL on the Messidor\ndataset, achieving an area under the precision-recall curve of 0.79 while\nutilizing only 76.54\\% of the available labeled data. These capabilities can\nenhances the practical usability of a DL model in the medical field, providing\na means of trust calibration in domain experts and a suitable solution for\nlearning in the data scarcity context often found.\n", "date": "2024-04-06", "categories": []}
{"id": "2412.19396", "title": "Comparing Few to Rank Many: Active Human Preference Learning using\n  Randomized Frank-Wolfe", "abstract": "  We study learning of human preferences from a limited comparison feedback.\nThis task is ubiquitous in machine learning. Its applications such as\nreinforcement learning from human feedback, have been transformational. We\nformulate this problem as learning a Plackett-Luce model over a universe of $N$\nchoices from $K$-way comparison feedback, where typically $K \\ll N$. Our\nsolution is the D-optimal design for the Plackett-Luce objective. The design\ndefines a data logging policy that elicits comparison feedback for a small\ncollection of optimally chosen points from all ${N \\choose K}$ feasible\nsubsets. The main algorithmic challenge in this work is that even fast methods\nfor solving D-optimal designs would have $O({N \\choose K})$ time complexity. To\naddress this issue, we propose a randomized Frank-Wolfe (FW) algorithm that\nsolves the linear maximization sub-problems in the FW method on randomly chosen\nvariables. We analyze the algorithm, and evaluate it empirically on synthetic\nand open-source NLP datasets.\n", "date": "2024-12-27", "categories": []}
{"id": "2412.17854", "title": "Active Geospatial Search for Efficient Tenant Eviction Outreach", "abstract": "  Tenant evictions threaten housing stability and are a major concern for many\ncities. An open question concerns whether data-driven methods enhance outreach\nprograms that target at-risk tenants to mitigate their risk of eviction. We\npropose a novel active geospatial search (AGS) modeling framework for this\nproblem. AGS integrates property-level information in a search policy that\nidentifies a sequence of rental units to canvas to both determine their\neviction risk and provide support if needed. We propose a hierarchical\nreinforcement learning approach to learn a search policy for AGS that scales to\nlarge urban areas containing thousands of parcels, balancing exploration and\nexploitation and accounting for travel costs and a budget constraint.\nCrucially, the search policy adapts online to newly discovered information\nabout evictions. Evaluation using eviction data for a large urban area\ndemonstrates that the proposed framework and algorithmic approach are\nconsiderably more effective at sequentially identifying eviction cases than\nbaseline methods.\n", "date": "2024-12-19", "categories": []}
{"id": "2412.16227", "title": "GALOT: Generative Active Learning via Optimizable Zero-shot\n  Text-to-image Generation", "abstract": "  Active Learning (AL) represents a crucial methodology within machine\nlearning, emphasizing the identification and utilization of the most\ninformative samples for efficient model training. However, a significant\nchallenge of AL is its dependence on the limited labeled data samples and data\ndistribution, resulting in limited performance. To address this limitation,\nthis paper integrates the zero-shot text-to-image (T2I) synthesis and active\nlearning by designing a novel framework that can efficiently train a machine\nlearning (ML) model sorely using the text description. Specifically, we\nleverage the AL criteria to optimize the text inputs for generating more\ninformative and diverse data samples, annotated by the pseudo-label crafted\nfrom text, then served as a synthetic dataset for active learning. This\napproach reduces the cost of data collection and annotation while increasing\nthe efficiency of model training by providing informative training samples,\nenabling a novel end-to-end ML task from text description to vision models.\nThrough comprehensive evaluations, our framework demonstrates consistent and\nsignificant improvements over traditional AL methods.\n", "date": "2024-12-18", "categories": []}
{"id": "2403.09259", "title": "To Label or Not to Label: Hybrid Active Learning for Neural Machine\n  Translation", "abstract": "  Active learning (AL) techniques reduce labeling costs for training neural\nmachine translation (NMT) models by selecting smaller representative subsets\nfrom unlabeled data for annotation. Diversity sampling techniques select\nheterogeneous instances, while uncertainty sampling methods select instances\nwith the highest model uncertainty. Both approaches have limitations -\ndiversity methods may extract varied but trivial examples, while uncertainty\nsampling can yield repetitive, uninformative instances. To bridge this gap, we\npropose Hybrid Uncertainty and Diversity Sampling (HUDS), an AL strategy for\ndomain adaptation in NMT that combines uncertainty and diversity for sentence\nselection. HUDS computes uncertainty scores for unlabeled sentences and\nsubsequently stratifies them. It then clusters sentence embeddings within each\nstratum and computes diversity scores by distance to the centroid. A weighted\nhybrid score that combines uncertainty and diversity is then used to select the\ntop instances for annotation in each AL iteration. Experiments on multi-domain\nGerman-English and French-English datasets demonstrate the better performance\nof HUDS over other strong AL baselines. We analyze the sentence selection with\nHUDS and show that it prioritizes diverse instances having high model\nuncertainty for annotation in early AL iterations.\n", "date": "2024-03-14", "categories": []}
{"id": "2310.05502", "title": "XAL: EXplainable Active Learning Makes Classifiers Better Low-resource\n  Learners", "abstract": "  Active learning (AL), which aims to construct an effective training set by\niteratively curating the most formative unlabeled data for annotation, has been\nwidely used in low-resource tasks. Most active learning techniques in\nclassification rely on the model's uncertainty or disagreement to choose\nunlabeled data, suffering from the problem of over-confidence in superficial\npatterns and a lack of exploration. Inspired by the cognitive processes in\nwhich humans deduce and predict through causal information, we take an initial\nattempt towards integrating rationales into AL and propose a novel Explainable\nActive Learning framework (XAL) for low-resource text classification, which\naims to encourage classifiers to justify their inferences and delve into\nunlabeled data for which they cannot provide reasonable explanations.\nSpecifically, besides using a pre-trained bi-directional encoder for\nclassification, we employ a pre-trained uni-directional decoder to generate and\nscore the explanation. We further facilitate the alignment of the model with\nhuman reasoning preference through a proposed ranking loss. During the\nselection of unlabeled data, the predicted uncertainty of the encoder and the\nexplanation score of the decoder complement each other as the final metric to\nacquire informative data. Extensive experiments on six datasets show that XAL\nachieves consistent improvement over 9 strong baselines. Analysis indicates\nthat the proposed method can generate corresponding explanations for its\npredictions.\n", "date": "2023-10-09", "categories": []}
{"id": "2311.17874", "title": "FisherRF: Active View Selection and Uncertainty Quantification for\n  Radiance Fields using Fisher Information", "abstract": "  This study addresses the challenging problem of active view selection and\nuncertainty quantification within the domain of Radiance Fields. Neural\nRadiance Fields (NeRF) have greatly advanced image rendering and\nreconstruction, but the cost of acquiring images poses the need to select the\nmost informative viewpoints efficiently. Existing approaches depend on\nmodifying the model architecture or hypothetical perturbation field to\nindirectly approximate the model uncertainty. However, selecting views from\nindirect approximation does not guarantee optimal information gain for the\nmodel. By leveraging Fisher Information, we directly quantify observed\ninformation on the parameters of Radiance Fields and select candidate views by\nmaximizing the Expected Information Gain(EIG). Our method achieves\nstate-of-the-art results on multiple tasks, including view selection, active\nmapping, and uncertainty quantification, demonstrating its potential to advance\nthe field of Radiance Fields.\n", "date": "2023-11-29", "categories": []}
{"id": "2410.13641", "title": "An Active Learning Framework for Inclusive Generation by Large Language\n  Models", "abstract": "  Ensuring that Large Language Models (LLMs) generate text representative of\ndiverse sub-populations is essential, particularly when key concepts related to\nunder-represented groups are scarce in the training data. We address this\nchallenge with a novel clustering-based active learning framework, enhanced\nwith knowledge distillation. The proposed framework transforms the intermediate\noutputs of the learner model, enabling effective active learning for generative\ntasks for the first time. Integration of clustering and knowledge distillation\nyields more representative models without prior knowledge of underlying data\ndistribution and overbearing human efforts. We validate our approach in\npractice through case studies in counter-narration and style transfer. We\nconstruct two new datasets in tandem with model training, showing a performance\nimprovement of 2%-10% over baseline models. Our results also show more\nconsistent performance across various data subgroups and increased lexical\ndiversity, underscoring our model's resilience to skewness in available data.\nFurther, our results show that the data acquired via our approach improves the\nperformance of secondary models not involved in the learning loop, showcasing\npractical utility of the framework.\n", "date": "2024-10-17", "categories": []}
{"id": "2412.11104", "title": "ABC3: Active Bayesian Causal Inference with Cohn Criteria in Randomized\n  Experiments", "abstract": "  In causal inference, randomized experiment is a de facto method to overcome\nvarious theoretical issues in observational study. However, the experimental\ndesign requires expensive costs, so an efficient experimental design is\nnecessary. We propose ABC3, a Bayesian active learning policy for causal\ninference. We show a policy minimizing an estimation error on conditional\naverage treatment effect is equivalent to minimizing an integrated posterior\nvariance, similar to Cohn criteria \\citep{cohn1994active}. We theoretically\nprove ABC3 also minimizes an imbalance between the treatment and control groups\nand the type 1 error probability. Imbalance-minimizing characteristic is\nespecially notable as several works have emphasized the importance of achieving\nbalance. Through extensive experiments on real-world data sets, ABC3 achieves\nthe highest efficiency, while empirically showing the theoretical results hold.\n", "date": "2024-12-15", "categories": []}
{"id": "2412.09701", "title": "CUAL: Continual Uncertainty-aware Active Learner", "abstract": "  AI deployed in many real-world use cases should be capable of adapting to\nnovelties encountered after deployment. Here, we consider a challenging,\nunder-explored and realistic continual adaptation problem: a deployed AI agent\nis continuously provided with unlabeled data that may contain not only unseen\nsamples of known classes but also samples from novel (unknown) classes. In such\na challenging setting, it has only a tiny labeling budget to query the most\ninformative samples to help it continuously learn. We present a comprehensive\nsolution to this complex problem with our model \"CUAL\" (Continual\nUncertainty-aware Active Learner). CUAL leverages an uncertainty estimation\nalgorithm to prioritize active labeling of ambiguous (uncertain) predicted\nnovel class samples while also simultaneously pseudo-labeling the most certain\npredictions of each class. Evaluations across multiple datasets, ablations,\nsettings and backbones (e.g. ViT foundation model) demonstrate our method's\neffectiveness. We will release our code upon acceptance.\n", "date": "2024-12-12", "categories": []}
{"id": "2412.08138", "title": "Learn How to Query from Unlabeled Data Streams in Federated Learning", "abstract": "  Federated learning (FL) enables collaborative learning among decentralized\nclients while safeguarding the privacy of their local data. Existing studies on\nFL typically assume offline labeled data available at each client when the\ntraining starts. Nevertheless, the training data in practice often arrive at\nclients in a streaming fashion without ground-truth labels. Given the expensive\nannotation cost, it is critical to identify a subset of informative samples for\nlabeling on clients. However, selecting samples locally while accommodating the\nglobal training objective presents a challenge unique to FL. In this work, we\ntackle this conundrum by framing the data querying process in FL as a\ncollaborative decentralized decision-making problem and proposing an effective\nsolution named LeaDQ, which leverages multi-agent reinforcement learning\nalgorithms. In particular, under the implicit guidance from global information,\nLeaDQ effectively learns the local policies for distributed clients and steers\nthem towards selecting samples that can enhance the global model's accuracy.\nExtensive simulations on image and text tasks show that LeaDQ advances the\nmodel performance in various FL scenarios, outperforming the benchmarking\nalgorithms.\n", "date": "2024-12-11", "categories": []}
{"id": "2412.09053", "title": "Safe Active Learning for Gaussian Differential Equations", "abstract": "  Gaussian Process differential equations (GPODE) have recently gained momentum\ndue to their ability to capture dynamics behavior of systems and also represent\nuncertainty in predictions. Prior work has described the process of training\nthe hyperparameters and, thereby, calibrating GPODE to data. How to design\nefficient algorithms to collect data for training GPODE models is still an open\nfield of research. Nevertheless high-quality training data is key for model\nperformance. Furthermore, data collection leads to time-cost and financial-cost\nand might in some areas even be safety critical to the system under test.\nTherefore, algorithms for safe and efficient data collection are central for\nbuilding high quality GPODE models. Our novel Safe Active Learning (SAL) for\nGPODE algorithm addresses this challenge by suggesting a mechanism to propose\nefficient and non-safety-critical data to collect. SAL GPODE does so by\nsequentially suggesting new data, measuring it and updating the GPODE model\nwith the new data. In this way, subsequent data points are iteratively\nsuggested. The core of our SAL GPODE algorithm is a constrained optimization\nproblem maximizing information of new data for GPODE model training constrained\nby the safety of the underlying system. We demonstrate our novel SAL GPODE's\nsuperiority compared to a standard, non-active way of measuring new data on two\nrelevant examples.\n", "date": "2024-12-12", "categories": []}
{"id": "2412.09126", "title": "Enhancing Modality Representation and Alignment for Multimodal\n  Cold-start Active Learning", "abstract": "  Training multimodal models requires a large amount of labeled data. Active\nlearning (AL) aim to reduce labeling costs. Most AL methods employ warm-start\napproaches, which rely on sufficient labeled data to train a well-calibrated\nmodel that can assess the uncertainty and diversity of unlabeled data. However,\nwhen assembling a dataset, labeled data are often scarce initially, leading to\na cold-start problem. Additionally, most AL methods seldom address multimodal\ndata, highlighting a research gap in this field. Our research addresses these\nissues by developing a two-stage method for Multi-Modal Cold-Start Active\nLearning (MMCSAL).\n  Firstly, we observe the modality gap, a significant distance between the\ncentroids of representations from different modalities, when only using\ncross-modal pairing information as self-supervision signals. This modality gap\naffects data selection process, as we calculate both uni-modal and cross-modal\ndistances. To address this, we introduce uni-modal prototypes to bridge the\nmodality gap. Secondly, conventional AL methods often falter in multimodal\nscenarios where alignment between modalities is overlooked. Therefore, we\npropose enhancing cross-modal alignment through regularization, thereby\nimproving the quality of selected multimodal data pairs in AL. Finally, our\nexperiments demonstrate MMCSAL's efficacy in selecting multimodal data pairs\nacross three multimodal datasets.\n", "date": "2024-12-12", "categories": []}
{"id": "2411.12010", "title": "Active learning for efficient discovery of optimal gene combinations in\n  the combinatorial perturbation space", "abstract": "  The advancement of novel combinatorial CRISPR screening technologies enables\nthe identification of synergistic gene combinations on a large scale. This is\ncrucial for developing novel and effective combination therapies, but the\ncombinatorial space makes exhaustive experimentation infeasible. We introduce\nNAIAD, an active learning framework that efficiently discovers optimal gene\npairs capable of driving cells toward desired cellular phenotypes. NAIAD\nleverages single-gene perturbation effects and adaptive gene embeddings that\nscale with the training data size, mitigating overfitting in small-sample\nlearning while capturing complex gene interactions as more data is collected.\nEvaluated on four CRISPR combinatorial perturbation datasets totaling over\n350,000 genetic interactions, NAIAD, trained on small datasets, outperforms\nexisting models by up to 40\\% relative to the second-best. NAIAD's\nrecommendation system prioritizes gene pairs with the maximum predicted\neffects, resulting in the highest marginal gain in each AI-experiment round and\naccelerating discovery with fewer CRISPR experimental iterations. Our NAIAD\nframework (https://github.com/NeptuneBio/NAIAD) improves the identification of\nnovel, effective gene combinations, enabling more efficient CRISPR library\ndesign and offering promising applications in genomics research and therapeutic\ndevelopment.\n", "date": "2024-11-18", "categories": []}
{"id": "2412.08225", "title": "Improving Active Learning with a Bayesian Representation of Epistemic\n  Uncertainty", "abstract": "  A popular strategy for active learning is to specifically target a reduction\nin epistemic uncertainty, since aleatoric uncertainty is often considered as\nbeing intrinsic to the system of interest and therefore not reducible. Yet,\ndistinguishing these two types of uncertainty remains challenging and there is\nno single strategy that consistently outperforms the others. We propose to use\na particular combination of probability and possibility theories, with the aim\nof using the latter to specifically represent epistemic uncertainty, and we\nshow how this combination leads to new active learning strategies that have\ndesirable properties. In order to demonstrate the efficiency of these\nstrategies in non-trivial settings, we introduce the notion of a possibilistic\nGaussian process (GP) and consider GP-based multiclass and binary\nclassification problems, for which the proposed methods display a strong\nperformance for both simulated and real datasets.\n", "date": "2024-12-11", "categories": []}
{"id": "2402.03235", "title": "ActiveAnno3D -- An Active Learning Framework for Multi-Modal 3D Object\n  Detection", "abstract": "  The curation of large-scale datasets is still costly and requires much time\nand resources. Data is often manually labeled, and the challenge of creating\nhigh-quality datasets remains. In this work, we fill the research gap using\nactive learning for multi-modal 3D object detection. We propose ActiveAnno3D,\nan active learning framework to select data samples for labeling that are of\nmaximum informativeness for training. We explore various continuous training\nmethods and integrate the most efficient method regarding computational demand\nand detection performance. Furthermore, we perform extensive experiments and\nablation studies with BEVFusion and PV-RCNN on the nuScenes and TUM Traffic\nIntersection dataset. We show that we can achieve almost the same performance\nwith PV-RCNN and the entropy-based query strategy when using only half of the\ntraining data (77.25 mAP compared to 83.50 mAP) of the TUM Traffic Intersection\ndataset. BEVFusion achieved an mAP of 64.31 when using half of the training\ndata and 75.0 mAP when using the complete nuScenes dataset. We integrate our\nactive learning framework into the proAnno labeling tool to enable AI-assisted\ndata selection and labeling and minimize the labeling costs. Finally, we\nprovide code, weights, and visualization results on our website:\nhttps://active3d-framework.github.io/active3d-framework.\n", "date": "2024-02-05", "categories": []}
{"id": "2412.06642", "title": "Class Balance Matters to Active Class-Incremental Learning", "abstract": "  Few-Shot Class-Incremental Learning has shown remarkable efficacy in\nefficient learning new concepts with limited annotations. Nevertheless, the\nheuristic few-shot annotations may not always cover the most informative\nsamples, which largely restricts the capability of incremental learner. We aim\nto start from a pool of large-scale unlabeled data and then annotate the most\ninformative samples for incremental learning. Based on this premise, this paper\nintroduces the Active Class-Incremental Learning (ACIL). The objective of ACIL\nis to select the most informative samples from the unlabeled pool to\neffectively train an incremental learner, aiming to maximize the performance of\nthe resulting model. Note that vanilla active learning algorithms suffer from\nclass-imbalanced distribution among annotated samples, which restricts the\nability of incremental learning. To achieve both class balance and\ninformativeness in chosen samples, we propose Class-Balanced Selection (CBS)\nstrategy. Specifically, we first cluster the features of all unlabeled images\ninto multiple groups. Then for each cluster, we employ greedy selection\nstrategy to ensure that the Gaussian distribution of the sampled features\nclosely matches the Gaussian distribution of all unlabeled features within the\ncluster. Our CBS can be plugged and played into those CIL methods which are\nbased on pretrained models with prompts tunning technique. Extensive\nexperiments under ACIL protocol across five diverse datasets demonstrate that\nCBS outperforms both random selection and other SOTA active learning\napproaches. Code is publicly available at https://github.com/1170300714/CBS.\n", "date": "2024-12-09", "categories": []}
{"id": "2302.00098", "title": "Does Deep Active Learning Work in the Wild?", "abstract": "  Deep active learning (DAL) methods have shown significant improvements in\nsample efficiency compared to simple random sampling. While these studies are\nvaluable, they nearly always assume that optimal DAL hyperparameter (HP)\nsettings are known in advance, or optimize the HPs through repeating DAL\nseveral times with different HP settings. Here, we argue that in real-world\nsettings, or in the wild, there is significant uncertainty regarding good HPs,\nand their optimization contradicts the premise of using DAL (i.e., we require\nlabeling efficiency). In this study, we evaluate the performance of eleven\nmodern DAL methods on eight benchmark problems as we vary a key HP shared by\nall methods: the pool ratio. Despite adjusting only one HP, our results\nindicate that eight of the eleven DAL methods sometimes underperform relative\nto simple random sampling and some frequently perform worse. Only three methods\nalways outperform random sampling (albeit narrowly), and we find that these\nmethods all utilize diversity to select samples - a relatively simple\ncriterion. Our findings reveal the limitations of existing DAL methods when\ndeployed in the wild, and present this as an important new open problem in the\nfield.\n", "date": "2023-01-31", "categories": []}
{"id": "2405.07440", "title": "Maximizing Information Gain in Privacy-Aware Active Learning of Email\n  Anomalies", "abstract": "  Redacted emails satisfy most privacy requirements but they make it more\ndifficult to detect anomalous emails that may be indicative of data\nexfiltration. In this paper we develop an enhanced method of Active Learning\nusing an information gain maximizing heuristic, and we evaluate its\neffectiveness in a real world setting where only redacted versions of email\ncould be labeled by human analysts due to privacy concerns. In the first case\nstudy we examined how Active Learning should be carried out. We found that\nmodel performance was best when a single highly skilled (in terms of the\nlabelling task) analyst provided the labels. In the second case study we used\nconfidence ratings to estimate the labeling uncertainty of analysts and then\nprioritized instances for labeling based on the expected information gain (the\ndifference between model uncertainty and analyst uncertainty) that would be\nprovided by labelling each instance. We found that the information maximization\ngain heuristic improved model performance over existing sampling methods for\nActive Learning. Based on the results obtained, we recommend that analysts\nshould be screened, and possibly trained, prior to implementation of Active\nLearning in cybersecurity applications. We also recommend that the information\ngain maximizing sample method (based on expert confidence) should be used in\nearly stages of Active Learning, providing that well-calibrated confidence can\nbe obtained. We also note that the expertise of analysts should be assessed\nprior to Active Learning, as we found that analysts with lower labelling skill\nhad poorly calibrated (over-) confidence in their labels.\n", "date": "2024-05-13", "categories": []}
{"id": "2412.03098", "title": "A surprisal oracle for when every layer counts", "abstract": "  Active Curriculum Language Modeling (ACLM; Hong et al., 2023) is a learner\ndirected approach to training a language model. We proposed the original\nversion of this process in our submission to the BabyLM 2023 task, and now we\npropose an updated ACLM process for the BabyLM 2024 task. ACLM involves an\niteratively- and dynamically-constructed curriculum informed over the training\nprocess by a model of uncertainty; other training items that are similarly\nuncertain to a least certain candidate item are prioritized. Our new process\nimproves the similarity model so that it is more dynamic, and we run ACLM over\nthe most successful model from the BabyLM 2023 task: ELC-BERT (Charpentier and\nSamuel, 2023). We find that while our models underperform on fine-grained\ngrammatical inferences, they outperform the BabyLM 2024 official base-lines on\ncommon-sense and world-knowledge tasks. We make our code available at https:\n//github.com/asayeed/ActiveBaby.\n", "date": "2024-12-04", "categories": []}
{"id": "2412.02310", "title": "Active Learning via Classifier Impact and Greedy Selection for\n  Interactive Image Retrieval", "abstract": "  Active Learning (AL) is a user-interactive approach aimed at reducing\nannotation costs by selecting the most crucial examples to label. Although AL\nhas been extensively studied for image classification tasks, the specific\nscenario of interactive image retrieval has received relatively little\nattention. This scenario presents unique characteristics, including an open-set\nand class-imbalanced binary classification, starting with very few labeled\nsamples. We introduce a novel batch-mode Active Learning framework named GAL\n(Greedy Active Learning) that better copes with this application. It\nincorporates a new acquisition function for sample selection that measures the\nimpact of each unlabeled sample on the classifier. We further embed this\nstrategy in a greedy selection approach, better exploiting the samples within\neach batch. We evaluate our framework with both linear (SVM) and non-linear\nMLP/Gaussian Process classifiers. For the Gaussian Process case, we show a\ntheoretical guarantee on the greedy approximation. Finally, we assess our\nperformance for the interactive content-based image retrieval task on several\nbenchmarks and demonstrate its superiority over existing approaches and common\nbaselines. Code is available at https://github.com/barleah/GreedyAL.\n", "date": "2024-12-03", "categories": []}
{"id": "2412.00401", "title": "PAL -- Parallel active learning for machine-learned potentials", "abstract": "  Constructing datasets representative of the target domain is essential for\ntraining effective machine learning models. Active learning (AL) is a promising\nmethod that iteratively extends training data to enhance model performance\nwhile minimizing data acquisition costs. However, current AL workflows often\nrequire human intervention and lack parallelism, leading to inefficiencies and\nunderutilization of modern computational resources. In this work, we introduce\nPAL, an automated, modular, and parallel active learning library that\nintegrates AL tasks and manages their execution and communication on shared-\nand distributed-memory systems using the Message Passing Interface (MPI). PAL\nprovides users with the flexibility to design and customize all components of\ntheir active learning scenarios, including machine learning models with\nuncertainty estimation, oracles for ground truth labeling, and strategies for\nexploring the target space. We demonstrate that PAL significantly reduces\ncomputational overhead and improves scalability, achieving substantial\nspeed-ups through asynchronous parallelization on CPU and GPU hardware.\nApplications of PAL to several real-world scenarios - including ground-state\nreactions in biomolecular systems, excited-state dynamics of molecules,\nsimulations of inorganic clusters, and thermo-fluid dynamics - illustrate its\neffectiveness in accelerating the development of machine learning models. Our\nresults show that PAL enables efficient utilization of high-performance\ncomputing resources in active learning workflows, fostering advancements in\nscientific research and engineering applications.\n", "date": "2024-11-30", "categories": []}
{"id": "2408.03144", "title": "Active Learning for Level Set Estimation Using Randomized Straddle\n  Algorithms", "abstract": "  Level set estimation (LSE), the problem of identifying the set of input\npoints where a function takes value above (or below) a given threshold, is\nimportant in practical applications. When the function is expensive-to-evaluate\nand black-box, the \\textit{straddle} algorithm, which is a representative\nheuristic for LSE based on Gaussian process models, and its extensions having\ntheoretical guarantees have been developed. However, many of existing methods\ninclude a confidence parameter $\\beta^{1/2}_t$ that must be specified by the\nuser, and methods that choose $\\beta^{1/2}_t$ heuristically do not provide\ntheoretical guarantees. In contrast, theoretically guaranteed values of\n$\\beta^{1/2}_t$ need to be increased depending on the number of iterations and\ncandidate points, and are conservative and not good for practical performance.\nIn this study, we propose a novel method, the \\textit{randomized straddle}\nalgorithm, in which $\\beta_t$ in the straddle algorithm is replaced by a random\nsample from the chi-squared distribution with two degrees of freedom. The\nconfidence parameter in the proposed method has the advantages of not needing\nadjustment, not depending on the number of iterations and candidate points, and\nnot being conservative. Furthermore, we show that the proposed method has\ntheoretical guarantees that depend on the sample complexity and the number of\niterations. Finally, we confirm the usefulness of the proposed method through\nnumerical experiments using synthetic and real data.\n", "date": "2024-08-06", "categories": []}
{"id": "2411.18060", "title": "ORIS: Online Active Learning Using Reinforcement Learning-based\n  Inclusive Sampling for Robust Streaming Analytics System", "abstract": "  Effective labeled data collection plays a critical role in developing and\nfine-tuning robust streaming analytics systems. However, continuously labeling\ndocuments to filter relevant information poses significant challenges like\nlimited labeling budget or lack of high-quality labels. There is a need for\nefficient human-in-the-loop machine learning (HITL-ML) design to improve\nstreaming analytics systems. One particular HITL- ML approach is online active\nlearning, which involves iteratively selecting a small set of the most\ninformative documents for labeling to enhance the ML model performance. The\nperformance of such algorithms can get affected due to human errors in\nlabeling. To address these challenges, we propose ORIS, a method to perform\nOnline active learning using Reinforcement learning-based Inclusive Sampling of\ndocuments for labeling. ORIS aims to create a novel Deep Q-Network-based\nstrategy to sample incoming documents that minimize human errors in labeling\nand enhance the ML model performance. We evaluate the ORIS method on emotion\nrecognition tasks, and it outperforms traditional baselines in terms of both\nhuman labeling performance and the ML model performance.\n", "date": "2024-11-27", "categories": []}
{"id": "2311.15335", "title": "TORE: Token Recycling in Vision Transformers for Efficient Active Visual\n  Exploration", "abstract": "  Active Visual Exploration (AVE) optimizes the utilization of robotic\nresources in real-world scenarios by sequentially selecting the most\ninformative observations. However, modern methods require a high computational\nbudget due to processing the same observations multiple times through the\nautoencoder transformers.\n  As a remedy, we introduce a novel approach to AVE called TOken REcycling\n(TORE). It divides the encoder into extractor and aggregator components. The\nextractor processes each observation separately, enabling the reuse of tokens\npassed to the aggregator. Moreover, to further reduce the computations, we\ndecrease the decoder to only one block.\n  Through extensive experiments, we demonstrate that TORE outperforms\nstate-of-the-art methods while reducing computational overhead by up to 90\\%.\n", "date": "2023-11-26", "categories": []}
{"id": "2411.17444", "title": "Maximally Separated Active Learning", "abstract": "  Active Learning aims to optimize performance while minimizing annotation\ncosts by selecting the most informative samples from an unlabelled pool.\nTraditional uncertainty sampling often leads to sampling bias by choosing\nsimilar uncertain samples. We propose an active learning method that utilizes\nfixed equiangular hyperspherical points as class prototypes, ensuring\nconsistent inter-class separation and robust feature representations. Our\napproach introduces Maximally Separated Active Learning (MSAL) for uncertainty\nsampling and a combined strategy (MSAL-D) for incorporating diversity. This\nmethod eliminates the need for costly clustering steps, while maintaining\ndiversity through hyperspherical uniformity. We demonstrate strong performance\nover existing active learning techniques across five benchmark datasets,\nhighlighting the method's effectiveness and integration ease. The code is\navailable on GitHub.\n", "date": "2024-11-26", "categories": []}
{"id": "2312.13927", "title": "On the Convergence of Loss and Uncertainty-based Active Learning\n  Algorithms", "abstract": "  We investigate the convergence rates and data sample sizes required for\ntraining a machine learning model using a stochastic gradient descent (SGD)\nalgorithm, where data points are sampled based on either their loss value or\nuncertainty value. These training methods are particularly relevant for active\nlearning and data subset selection problems. For SGD with a constant step size\nupdate, we present convergence results for linear classifiers and linearly\nseparable datasets using squared hinge loss and similar training loss\nfunctions. Additionally, we extend our analysis to more general classifiers and\ndatasets, considering a wide range of loss-based sampling strategies and smooth\nconvex training loss functions. We propose a novel algorithm called\nAdaptive-Weight Sampling (AWS) that utilizes SGD with an adaptive step size\nthat achieves stochastic Polyak's step size in expectation. We establish\nconvergence rate results for AWS for smooth convex training loss functions. Our\nnumerical experiments demonstrate the efficiency of AWS on various datasets by\nusing either exact or estimated loss values.\n", "date": "2023-12-21", "categories": []}
{"id": "2411.15217", "title": "LPLgrad: Optimizing Active Learning Through Gradient Norm Sample\n  Selection and Auxiliary Model Training", "abstract": "  Machine learning models are increasingly being utilized across various fields\nand tasks due to their outstanding performance and strong generalization\ncapabilities. Nonetheless, their success hinges on the availability of large\nvolumes of annotated data, the creation of which is often labor-intensive,\ntime-consuming, and expensive. Many active learning (AL) approaches have been\nproposed to address these challenges, but they often fail to fully leverage the\ninformation from the core phases of AL, such as training on the labeled set and\nquerying new unlabeled samples. To bridge this gap, we propose a novel AL\napproach, Loss Prediction Loss with Gradient Norm (LPLgrad), designed to\nquantify model uncertainty effectively and improve the accuracy of image\nclassification tasks. LPLgrad operates in two distinct phases: (i) {\\em\nTraining Phase} aims to predict the loss for input features by jointly training\na main model and an auxiliary model. Both models are trained on the labeled\ndata to maximize the efficiency of the learning process, an aspect often\noverlooked in previous AL methods. This dual-model approach enhances the\nability to extract complex input features and learn intrinsic patterns from the\ndata effectively; (ii) {\\em Querying Phase} that quantifies the uncertainty of\nthe main model to guide sample selection. This is achieved by calculating the\ngradient norm of the entropy values for samples in the unlabeled dataset.\nSamples with the highest gradient norms are prioritized for labeling and\nsubsequently added to the labeled set, improving the model's performance with\nminimal labeling effort. Extensive evaluations on real-world datasets\ndemonstrate that the LPLgrad approach outperforms state-of-the-art methods by\norder of magnitude in terms of accuracy on a small number of labeled images,\nyet achieving comparable training and querying times in multiple image\nclassification tasks.\n", "date": "2024-11-20", "categories": []}
{"id": "2411.15763", "title": "Integrating Deep Metric Learning with Coreset for Active Learning in 3D\n  Segmentation", "abstract": "  Deep learning has seen remarkable advancements in machine learning, yet it\noften demands extensive annotated data. Tasks like 3D semantic segmentation\nimpose a substantial annotation burden, especially in domains like medicine,\nwhere expert annotations drive up the cost. Active learning (AL) holds great\npotential to alleviate this annotation burden in 3D medical segmentation. The\nmajority of existing AL methods, however, are not tailored to the medical\ndomain. While weakly-supervised methods have been explored to reduce annotation\nburden, the fusion of AL with weak supervision remains unexplored, despite its\npotential to significantly reduce annotation costs. Additionally, there is\nlittle focus on slice-based AL for 3D segmentation, which can also\nsignificantly reduce costs in comparison to conventional volume-based AL. This\npaper introduces a novel metric learning method for Coreset to perform\nslice-based active learning in 3D medical segmentation. By merging contrastive\nlearning with inherent data groupings in medical imaging, we learn a metric\nthat emphasizes the relevant differences in samples for training 3D medical\nsegmentation models. We perform comprehensive evaluations using both weak and\nfull annotations across four datasets (medical and non-medical). Our findings\ndemonstrate that our approach surpasses existing active learning techniques on\nboth weak and full annotations and obtains superior performance with\nlow-annotation budgets which is crucial in medical imaging. Source code for\nthis project is available in the supplementary materials and on GitHub:\nhttps://github.com/arvindmvepa/al-seg.\n", "date": "2024-11-24", "categories": []}
{"id": "2411.01925", "title": "Exploiting Contextual Uncertainty of Visual Data for Efficient Training\n  of Deep Models", "abstract": "  Objects, in the real world, rarely occur in isolation and exhibit typical\narrangements governed by their independent utility, and their expected\ninteraction with humans and other objects in the context. For example, a chair\nis expected near a table, and a computer is expected on top. Humans use this\nspatial context and relative placement as an important cue for visual\nrecognition in case of ambiguities. Similar to human's, DNN's exploit\ncontextual information from data to learn representations. Our research focuses\non harnessing the contextual aspects of visual data to optimize data annotation\nand enhance the training of deep networks. Our contributions can be summarized\nas follows: (1) We introduce the notion of contextual diversity for active\nlearning CDAL and show its applicability in three different visual tasks\nsemantic segmentation, object detection and image classification, (2) We\npropose a data repair algorithm to curate contextually fair data to reduce\nmodel bias, enabling the model to detect objects out of their obvious context,\n(3) We propose Class-based annotation, where contextually relevant classes are\nselected that are complementary for model training under domain shift.\nUnderstanding the importance of well-curated data, we also emphasize the\nnecessity of involving humans in the loop to achieve accurate annotations and\nto develop novel interaction strategies that allow humans to serve as\nfact-checkers. In line with this we are working on developing image retrieval\nsystem for wildlife camera trap images and reliable warning system for poor\nquality rural roads. For large-scale annotation, we are employing a strategic\ncombination of human expertise and zero-shot models, while also integrating\nhuman input at various stages for continuous feedback.\n", "date": "2024-11-04", "categories": []}
{"id": "2411.12563", "title": "Stream-Based Active Learning for Process Monitoring", "abstract": "  Statistical process monitoring (SPM) methods are essential tools in quality\nmanagement to check the stability of industrial processes, i.e., to dynamically\nclassify the process state as in control (IC), under normal operating\nconditions, or out of control (OC), otherwise. Traditional SPM methods are\nbased on unsupervised approaches, which are popular because in most industrial\napplications the true OC states of the process are not explicitly known. This\nhampered the development of supervised methods that could instead take\nadvantage of process data containing labels on the true process state, although\nthey still need improvement in dealing with class imbalance, as OC states are\nrare in high-quality processes, and the dynamic recognition of unseen classes,\ne.g., the number of possible OC states. This article presents a novel\nstream-based active learning strategy for SPM that enhances partially hidden\nMarkov models to deal with data streams. The ultimate goal is to optimize\nlabeling resources constrained by a limited budget and dynamically update the\npossible OC states. The proposed method performance in classifying the true\nstate of the process is assessed through a simulation and a case study on the\nSPM of a resistance spot welding process in the automotive industry, which\nmotivated this research.\n", "date": "2024-11-19", "categories": []}
{"id": "2403.01101", "title": "Feature Alignment: Rethinking Efficient Active Learning via Proxy in the\n  Context of Pre-trained Models", "abstract": "  Fine-tuning the pre-trained model with active learning holds promise for\nreducing annotation costs. However, this combination introduces significant\ncomputational costs, particularly with the growing scale of pre-trained models.\nRecent research has proposed proxy-based active learning, which pre-computes\nfeatures to reduce computational costs. Yet, this approach often incurs a\nsignificant loss in active learning performance, sometimes outweighing the\ncomputational cost savings. This paper demonstrates that not all sample\nselection differences result in performance degradation. Furthermore, we show\nthat suitable training methods can mitigate the decline of active learning\nperformance caused by certain selection discrepancies. Building upon detailed\nanalysis, we propose a novel method, aligned selection via proxy, which\nimproves proxy-based active learning performance by updating pre-computed\nfeatures and selecting a proper training method. Extensive experiments validate\nthat our method improves the total cost of efficient active learning while\nmaintaining computational efficiency. The code is available at\n\\url{https://github.com/ZiTingW/asvp}.\n", "date": "2024-03-02", "categories": []}
{"id": "2411.10601", "title": "Learning Quantitative Automata Modulo Theories", "abstract": "  Quantitative automata are useful representations for numerous applications,\nincluding modeling probability distributions over sequences to Markov chains\nand reward machines. Actively learning such automata typically occurs using\nexplicitly gathered input-output examples under adaptations of the L-star\nalgorithm. However, obtaining explicit input-output pairs can be expensive, and\nthere exist scenarios, including preference-based learning or learning from\nrankings, where providing constraints is a less exerting and a more natural way\nto concisely describe desired properties. Consequently, we propose the problem\nof learning deterministic quantitative automata from sets of constraints over\nthe valuations of input sequences. We present QUINTIC, an active learning\nalgorithm, wherein the learner infers a valid automaton through deductive\nreasoning, by applying a theory to a set of currently available constraints and\nan assumed preference model and quantitative automaton class. QUINTIC performs\na complete search over the space of automata, and is guaranteed to be minimal\nand correctly terminate. Our evaluations utilize theory of rationals in order\nto learn summation, discounted summation, product, and classification\nquantitative automata, and indicate QUINTIC is effective at learning these\ntypes of automata.\n", "date": "2024-11-15", "categories": []}
{"id": "2411.10896", "title": "Targeting Negative Flips in Active Learning using Validation Sets", "abstract": "  The performance of active learning algorithms can be improved in two ways.\nThe often used and intuitive way is by reducing the overall error rate within\nthe test set. The second way is to ensure that correct predictions are not\nforgotten when the training set is increased in between rounds. The former is\nmeasured by the accuracy of the model and the latter is captured in negative\nflips between rounds. Negative flips are samples that are correctly predicted\nwhen trained with the previous/smaller dataset and incorrectly predicted after\nadditional samples are labeled. In this paper, we discuss improving the\nperformance of active learning algorithms both in terms of prediction accuracy\nand negative flips. The first observation we make in this paper is that\nnegative flips and overall error rates are decoupled and reducing one does not\nnecessarily imply that the other is reduced. Our observation is important as\ncurrent active learning algorithms do not consider negative flips directly and\nimplicitly assume the opposite. The second observation is that performing\ntargeted active learning on subsets of the unlabeled pool has a significant\nimpact on the behavior of the active learning algorithm and influences both\nnegative flips and prediction accuracy. We then develop ROSE - a plug-in\nalgorithm that utilizes a small labeled validation set to restrict arbitrary\nactive learning acquisition functions to negative flips within the unlabeled\npool. We show that integrating a validation set results in a significant\nperformance boost in terms of accuracy, negative flip rate reduction, or both.\n", "date": "2024-11-16", "categories": []}
{"id": "2411.11256", "title": "Progressive Generalization Risk Reduction for Data-Efficient Causal\n  Effect Estimation", "abstract": "  Causal effect estimation (CEE) provides a crucial tool for predicting the\nunobserved counterfactual outcome for an entity. As CEE relaxes the requirement\nfor ``perfect'' counterfactual samples (e.g., patients with identical\nattributes and only differ in treatments received) that are impractical to\nobtain and can instead operate on observational data, it is usually used in\nhigh-stake domains like medical treatment effect prediction. Nevertheless, in\nthose high-stake domains, gathering a decently sized, fully labelled\nobservational dataset remains challenging due to hurdles associated with costs,\nethics, expertise and time needed, etc., of which medical treatment surveys are\na typical example. Consequently, if the training dataset is small in scale, low\ngeneralization risks can hardly be achieved on any CEE algorithms.\n  Unlike existing CEE methods that assume the constant availability of a\ndataset with abundant samples, in this paper, we study a more realistic CEE\nsetting where the labelled data samples are scarce at the beginning, while more\ncan be gradually acquired over the course of training -- assuredly under a\nlimited budget considering their expensive nature. Then, the problem naturally\ncomes down to actively selecting the best possible samples to be labelled,\ne.g., identifying the next subset of patients to conduct the treatment survey.\nHowever, acquiring quality data for reducing the CEE risk under limited\nlabelling budgets remains under-explored until now. To fill the gap, we\ntheoretically analyse the generalization risk from an intriguing perspective of\nprogressively shrinking its upper bound, and develop a principled label\nacquisition pipeline exclusively for CEE tasks. With our analysis, we propose\nthe Model Agnostic Causal Active Learning (MACAL) algorithm for batch-wise\nlabel acquisition, which aims to reduce both the CEE model's uncertainty and\nthe post-acquisition ...\n", "date": "2024-11-18", "categories": []}
{"id": "2410.17142", "title": "Coniferest: a complete active anomaly detection framework", "abstract": "  We present coniferest, an open source generic purpose active anomaly\ndetection framework written in Python. The package design and implemented\nalgorithms are described. Currently, static outlier detection analysis is\nsupported via the Isolation forest algorithm. Moreover, Active Anomaly\nDiscovery (AAD) and Pineforest algorithms are available to tackle active\nanomaly detection problems. The algorithms and package performance are\nevaluated on a series of synthetic datasets. We also describe a few success\ncases which resulted from applying the package to real astronomical data in\nactive anomaly detection tasks within the SNAD project.\n", "date": "2024-10-22", "categories": []}
{"id": "2402.15198", "title": "Bidirectional Uncertainty-Based Active Learning for Open Set Annotation", "abstract": "  Active learning (AL) in open set scenarios presents a novel challenge of\nidentifying the most valuable examples in an unlabeled data pool that comprises\ndata from both known and unknown classes. Traditional methods prioritize\nselecting informative examples with low confidence, with the risk of mistakenly\nselecting unknown-class examples with similarly low confidence. Recent methods\nfavor the most probable known-class examples, with the risk of picking simple\nalready mastered examples. In this paper, we attempt to query examples that are\nboth likely from known classes and highly informative, and propose a\nBidirectional Uncertainty-based Active Learning (BUAL) framework. Specifically,\nwe achieve this by first pushing the unknown class examples toward regions with\nhigh-confidence predictions, i.e., the proposed Random Label Negative Learning\nmethod. Then, we propose a Bidirectional Uncertainty sampling strategy by\njointly estimating uncertainty posed by both positive and negative learning to\nperform consistent and stable sampling. BUAL successfully extends existing\nuncertainty-based AL methods to complex open-set scenarios. Extensive\nexperiments on multiple datasets with varying openness demonstrate that BUAL\nachieves state-of-the-art performance. The code is available at\nhttps://github.com/chenchenzong/BUAL.\n", "date": "2024-02-23", "categories": []}
{"id": "2408.00426", "title": "A Cross-Domain Benchmark for Active Learning", "abstract": "  Active Learning (AL) deals with identifying the most informative samples for\nlabeling to reduce data annotation costs for supervised learning tasks. AL\nresearch suffers from the fact that lifts from literature generalize poorly and\nthat only a small number of repetitions of experiments are conducted. To\novercome these obstacles, we propose CDALBench, the first active learning\nbenchmark which includes tasks in computer vision, natural language processing\nand tabular learning. Furthermore, by providing an efficient, greedy oracle,\nCDALBench can be evaluated with 50 runs for each experiment. We show, that both\nthe cross-domain character and a large amount of repetitions are crucial for\nsophisticated evaluation of AL research. Concretely, we show that the\nsuperiority of specific methods varies over the different domains, making it\nimportant to evaluate Active Learning with a cross-domain benchmark.\nAdditionally, we show that having a large amount of runs is crucial. With only\nconducting three runs as often done in the literature, the superiority of\nspecific methods can strongly vary with the specific runs. This effect is so\nstrong, that, depending on the seed, even a well-established method's\nperformance can be significantly better and significantly worse than random for\nthe same dataset.\n", "date": "2024-08-01", "categories": []}
{"id": "2411.05939", "title": "GCI-ViTAL: Gradual Confidence Improvement with Vision Transformers for\n  Active Learning on Label Noise", "abstract": "  Active learning aims to train accurate classifiers while minimizing labeling\ncosts by strategically selecting informative samples for annotation. This study\nfocuses on image classification tasks, comparing AL methods on CIFAR10,\nCIFAR100, Food101, and the Chest X-ray datasets under varying label noise\nrates. We investigate the impact of model architecture by comparing\nConvolutional Neural Networks (CNNs) and Vision Transformer (ViT)-based models.\nAdditionally, we propose a novel deep active learning algorithm, GCI-ViTAL,\ndesigned to be robust to label noise. GCI-ViTAL utilizes prediction entropy and\nthe Frobenius norm of last-layer attention vectors compared to class-centric\nclean set attention vectors. Our method identifies samples that are both\nuncertain and semantically divergent from typical images in their assigned\nclass. This allows GCI-ViTAL to select informative data points even in the\npresence of label noise while flagging potentially mislabeled candidates. Label\nsmoothing is applied to train a model that is not overly confident about\npotentially noisy labels. We evaluate GCI-ViTAL under varying levels of\nsymmetric label noise and compare it to five other AL strategies. Our results\ndemonstrate that using ViTs leads to significant performance improvements over\nCNNs across all AL strategies, particularly in noisy label settings. We also\nfind that using the semantic information of images as label grounding helps in\ntraining a more robust model under label noise. Notably, we do not perform\nextensive hyperparameter tuning, providing an out-of-the-box comparison that\naddresses the common challenge practitioners face in selecting models and\nactive learning strategies without an exhaustive literature review on training\nand fine-tuning vision models on real-world application data.\n", "date": "2024-11-08", "categories": []}
{"id": "2411.05752", "title": "FisherMask: Enhancing Neural Network Labeling Efficiency in Image\n  Classification Using Fisher Information", "abstract": "  Deep learning (DL) models are popular across various domains due to their\nremarkable performance and efficiency. However, their effectiveness relies\nheavily on large amounts of labeled data, which are often time-consuming and\nlabor-intensive to generate manually. To overcome this challenge, it is\nessential to develop strategies that reduce reliance on extensive labeled data\nwhile preserving model performance. In this paper, we propose FisherMask, a\nFisher information-based active learning (AL) approach that identifies key\nnetwork parameters by masking them based on their Fisher information values.\nFisherMask enhances batch AL by using Fisher information to select the most\ncritical parameters, allowing the identification of the most impactful samples\nduring AL training. Moreover, Fisher information possesses favorable\nstatistical properties, offering valuable insights into model behavior and\nproviding a better understanding of the performance characteristics within the\nAL pipeline. Our extensive experiments demonstrate that FisherMask\nsignificantly outperforms state-of-the-art methods on diverse datasets,\nincluding CIFAR-10 and FashionMNIST, especially under imbalanced settings.\nThese improvements lead to substantial gains in labeling efficiency. Hence\nserving as an effective tool to measure the sensitivity of model parameters to\ndata samples. Our code is available on\n\\url{https://github.com/sgchr273/FisherMask}.\n", "date": "2024-11-08", "categories": []}
{"id": "2408.07941", "title": "Robust Offline Active Learning on Graphs", "abstract": "  We consider the problem of active learning on graphs, which has crucial\napplications in many real-world networks where labeling node responses is\nexpensive. In this paper, we propose an offline active learning method that\nselects nodes to query by explicitly incorporating information from both the\nnetwork structure and node covariates. Building on graph signal recovery\ntheories and the random spectral sparsification technique, the proposed method\nadopts a two-stage biased sampling strategy that takes both informativeness and\nrepresentativeness into consideration for node querying. Informativeness refers\nto the complexity of graph signals that are learnable from the responses of\nqueried nodes, while representativeness refers to the capacity of queried nodes\nto control generalization errors given noisy node-level information. We\nestablish a theoretical relationship between generalization error and the\nnumber of nodes selected by the proposed method. Our theoretical results\ndemonstrate the trade-off between informativeness and representativeness in\nactive learning. Extensive numerical experiments show that the proposed method\nis competitive with existing graph-based active learning methods, especially\nwhen node covariates and responses contain noises. Additionally, the proposed\nmethod is applicable to both regression and classification tasks on graphs.\n", "date": "2024-08-15", "categories": []}
{"id": "2406.03845", "title": "Open Problem: Active Representation Learning", "abstract": "  In this work, we introduce the concept of Active Representation Learning, a\nnovel class of problems that intertwines exploration and representation\nlearning within partially observable environments. We extend ideas from Active\nSimultaneous Localization and Mapping (active SLAM), and translate them to\nscientific discovery problems, exemplified by adaptive microscopy. We explore\nthe need for a framework that derives exploration skills from representations\nthat are in some sense actionable, aiming to enhance the efficiency and\neffectiveness of data collection and model building in the natural sciences.\n", "date": "2024-06-06", "categories": []}
{"id": "2406.18621", "title": "Towards Deep Active Learning in Avian Bioacoustics", "abstract": "  Passive acoustic monitoring (PAM) in avian bioacoustics enables\ncost-effective and extensive data collection with minimal disruption to natural\nhabitats. Despite advancements in computational avian bioacoustics, deep\nlearning models continue to encounter challenges in adapting to diverse\nenvironments in practical PAM scenarios. This is primarily due to the scarcity\nof annotations, which requires labor-intensive efforts from human experts.\nActive learning (AL) reduces annotation cost and speed ups adaption to diverse\nscenarios by querying the most informative instances for labeling. This paper\noutlines a deep AL approach, introduces key challenges, and conducts a\nsmall-scale pilot study.\n", "date": "2024-06-26", "categories": []}
{"id": "2410.03918", "title": "STONE: A Submodular Optimization Framework for Active 3D Object\n  Detection", "abstract": "  3D object detection is fundamentally important for various emerging\napplications, including autonomous driving and robotics. A key requirement for\ntraining an accurate 3D object detector is the availability of a large amount\nof LiDAR-based point cloud data. Unfortunately, labeling point cloud data is\nextremely challenging, as accurate 3D bounding boxes and semantic labels are\nrequired for each potential object. This paper proposes a unified active 3D\nobject detection framework, for greatly reducing the labeling cost of training\n3D object detectors. Our framework is based on a novel formulation of\nsubmodular optimization, specifically tailored to the problem of active 3D\nobject detection. In particular, we address two fundamental challenges\nassociated with active 3D object detection: data imbalance and the need to\ncover the distribution of the data, including LiDAR-based point cloud data of\nvarying difficulty levels. Extensive experiments demonstrate that our method\nachieves state-of-the-art performance with high computational efficiency\ncompared to existing active learning methods. The code is available at\nhttps://github.com/RuiyuM/STONE.\n", "date": "2024-10-04", "categories": []}
{"id": "2411.00137", "title": "Cost-Aware Query Policies in Active Learning for Efficient Autonomous\n  Robotic Exploration", "abstract": "  In missions constrained by finite resources, efficient data collection is\ncritical. Informative path planning, driven by automated decision-making,\noptimizes exploration by reducing the costs associated with accurate\ncharacterization of a target in an environment. Previous implementations of\nactive learning did not consider the action cost for regression problems or\nonly considered the action cost for classification problems. This paper\nanalyzes an AL algorithm for Gaussian Process regression while incorporating\naction cost. The algorithm's performance is compared on various regression\nproblems to include terrain mapping on diverse simulated surfaces along metrics\nof root mean square error, samples and distance until convergence, and model\nvariance upon convergence. The cost-dependent acquisition policy doesn't\norganically optimize information gain over distance. Instead, the traditional\nuncertainty metric with a distance constraint best minimizes root-mean-square\nerror over trajectory distance. This studys impact is to provide insight into\nincorporating action cost with AL methods to optimize exploration under\nrealistic mission constraints.\n", "date": "2024-10-31", "categories": []}
{"id": "2410.22187", "title": "Active Learning for Vision-Language Models", "abstract": "  Pre-trained vision-language models (VLMs) like CLIP have demonstrated\nimpressive zero-shot performance on a wide range of downstream computer vision\ntasks. However, there still exists a considerable performance gap between these\nmodels and a supervised deep model trained on a downstream dataset. To bridge\nthis gap, we propose a novel active learning (AL) framework that enhances the\nzero-shot classification performance of VLMs by selecting only a few\ninformative samples from the unlabeled data for annotation during training. To\nachieve this, our approach first calibrates the predicted entropy of VLMs and\nthen utilizes a combination of self-uncertainty and neighbor-aware uncertainty\nto calculate a reliable uncertainty measure for active sample selection. Our\nextensive experiments show that the proposed approach outperforms existing AL\napproaches on several image classification datasets, and significantly enhances\nthe zero-shot performance of VLMs.\n", "date": "2024-10-29", "categories": []}
{"id": "2405.03059", "title": "Active Preference Learning for Ordering Items In- and Out-of-sample", "abstract": "  Learning an ordering of items based on pairwise comparisons is useful when\nitems are difficult to rate consistently on an absolute scale, for example,\nwhen annotators have to make subjective assessments. When exhaustive comparison\nis infeasible, actively sampling item pairs can reduce the number of\nannotations necessary for learning an accurate ordering. However, many\nalgorithms ignore shared structure between items, limiting their sample\nefficiency and precluding generalization to new items. It is also common to\ndisregard how noise in comparisons varies between item pairs, despite it being\ninformative of item similarity. In this work, we study active preference\nlearning for ordering items with contextual attributes, both in- and\nout-of-sample. We give an upper bound on the expected ordering error of a\nlogistic preference model as a function of which items have been compared.\nNext, we propose an active learning strategy that samples items to minimize\nthis bound by accounting for aleatoric and epistemic uncertainty in\ncomparisons. We evaluate the resulting algorithm, and a variant aimed at\nreducing model misspecification, in multiple realistic ordering tasks with\ncomparisons made by human annotators. Our results demonstrate superior sample\nefficiency and generalization compared to non-contextual ranking approaches and\nactive preference learning baselines.\n", "date": "2024-05-05", "categories": []}
{"id": "2407.05973", "title": "Active Label Refinement for Robust Training of Imbalanced Medical Image\n  Classification Tasks in the Presence of High Label Noise", "abstract": "  The robustness of supervised deep learning-based medical image classification\nis significantly undermined by label noise. Although several methods have been\nproposed to enhance classification performance in the presence of noisy labels,\nthey face some challenges: 1) a struggle with class-imbalanced datasets,\nleading to the frequent overlooking of minority classes as noisy samples; 2) a\nsingular focus on maximizing performance using noisy datasets, without\nincorporating experts-in-the-loop for actively cleaning the noisy labels. To\nmitigate these challenges, we propose a two-phase approach that combines\nLearning with Noisy Labels (LNL) and active learning. This approach not only\nimproves the robustness of medical image classification in the presence of\nnoisy labels, but also iteratively improves the quality of the dataset by\nrelabeling the important incorrect labels, under a limited annotation budget.\nFurthermore, we introduce a novel Variance of Gradients approach in LNL phase,\nwhich complements the loss-based sample selection by also sampling\nunder-represented samples. Using two imbalanced noisy medical classification\ndatasets, we demonstrate that that our proposed technique is superior to its\npredecessors at handling class imbalance by not misidentifying clean samples\nfrom minority classes as mostly noisy samples.\n", "date": "2024-07-08", "categories": []}
{"id": "2410.19176", "title": "Perturbation-based Graph Active Learning for Weakly-Supervised Belief\n  Representation Learning", "abstract": "  This paper addresses the problem of optimizing the allocation of labeling\nresources for semi-supervised belief representation learning in social\nnetworks. The objective is to strategically identify valuable messages on\nsocial media graphs that are worth labeling within a constrained budget,\nultimately maximizing the task's performance. Despite the progress in\nunsupervised or semi-supervised methods in advancing belief and ideology\nrepresentation learning on social networks and the remarkable efficacy of graph\nlearning techniques, the availability of high-quality curated labeled social\ndata can greatly benefit and further improve performances. Consequently,\nallocating labeling efforts is a critical research problem in scenarios where\nlabeling resources are limited. This paper proposes a graph data\naugmentation-inspired perturbation-based active learning strategy (PerbALGraph)\nthat progressively selects messages for labeling according to an automatic\nestimator, obviating human guidance. This estimator is based on the principle\nthat messages in the network that exhibit heightened sensitivity to structural\nfeatures of the observational data indicate landmark quality that significantly\ninfluences semi-supervision processes. We design the estimator to be the\nprediction variance under a set of designed graph perturbations, which is\nmodel-agnostic and application-independent. Extensive experiment results\ndemonstrate the effectiveness of the proposed strategy for belief\nrepresentation learning tasks.\n", "date": "2024-10-24", "categories": []}
{"id": "2404.01812", "title": "Uncertainty-aware Active Learning of NeRF-based Object Models for Robot\n  Manipulators using Visual and Re-orientation Actions", "abstract": "  Manipulating unseen objects is challenging without a 3D representation, as\nobjects generally have occluded surfaces. This requires physical interaction\nwith objects to build their internal representations. This paper presents an\napproach that enables a robot to rapidly learn the complete 3D model of a given\nobject for manipulation in unfamiliar orientations. We use an ensemble of\npartially constructed NeRF models to quantify model uncertainty to determine\nthe next action (a visual or re-orientation action) by optimizing\ninformativeness and feasibility. Further, our approach determines when and how\nto grasp and re-orient an object given its partial NeRF model and re-estimates\nthe object pose to rectify misalignments introduced during the interaction.\nExperiments with a simulated Franka Emika Robot Manipulator operating in a\ntabletop environment with benchmark objects demonstrate an improvement of (i)\n14% in visual reconstruction quality (PSNR), (ii) 20% in the geometric/depth\nreconstruction of the object surface (F-score) and (iii) 71% in the task\nsuccess rate of manipulating objects a-priori unseen orientations/stable\nconfigurations in the scene; over current methods. The project page can be\nfound here: https://actnerf.github.io.\n", "date": "2024-04-02", "categories": []}
{"id": "2404.15720", "title": "Annotator-Centric Active Learning for Subjective NLP Tasks", "abstract": "  Active Learning (AL) addresses the high costs of collecting human annotations\nby strategically annotating the most informative samples. However, for\nsubjective NLP tasks, incorporating a wide range of perspectives in the\nannotation process is crucial to capture the variability in human judgments. We\nintroduce Annotator-Centric Active Learning (ACAL), which incorporates an\nannotator selection strategy following data sampling. Our objective is\ntwo-fold: 1) to efficiently approximate the full diversity of human judgments,\nand 2) to assess model performance using annotator-centric metrics, which value\nminority and majority perspectives equally. We experiment with multiple\nannotator selection strategies across seven subjective NLP tasks, employing\nboth traditional and novel, human-centered evaluation metrics. Our findings\nindicate that ACAL improves data efficiency and excels in annotator-centric\nperformance evaluations. However, its success depends on the availability of a\nsufficiently large and diverse pool of annotators to sample from.\n", "date": "2024-04-24", "categories": []}
{"id": "2410.17917", "title": "regAL: Python Package for Active Learning of Regression Problems", "abstract": "  Increasingly more research areas rely on machine learning methods to\naccelerate discovery while saving resources. Machine learning models, however,\nusually require large datasets of experimental or computational results, which\nin certain fields, such as (bio)chemistry, materials science, or medicine, are\nrarely given and often prohibitively expensive to obtain. To bypass that\nobstacle, active learning methods are employed to develop machine learning\nmodels with a desired performance while requiring the least possible number of\ncomputational or experimental results from the domain of application. For this\npurpose, the model's knowledge about certain regions of the application domain\nis estimated to guide the choice of the model's training set. Although active\nlearning is widely studied for classification problems (discrete outcomes),\ncomparatively few works handle this method for regression problems (continuous\noutcomes). In this work, we present our Python package regAL, which allows\nusers to evaluate different active learning strategies for regression problems.\nWith a minimal input of just the dataset in question, but many additional\ncustomization and insight options, this package is intended for anyone who aims\nto perform and understand active learning in their problem-specific scope.\n", "date": "2024-10-23", "categories": []}
{"id": "2405.00452", "title": "Predictive Accuracy-Based Active Learning for Medical Image Segmentation", "abstract": "  Active learning is considered a viable solution to alleviate the\ncontradiction between the high dependency of deep learning-based segmentation\nmethods on annotated data and the expensive pixel-level annotation cost of\nmedical images. However, most existing methods suffer from unreliable\nuncertainty assessment and the struggle to balance diversity and\ninformativeness, leading to poor performance in segmentation tasks. In\nresponse, we propose an efficient Predictive Accuracy-based Active Learning\n(PAAL) method for medical image segmentation, first introducing predictive\naccuracy to define uncertainty. Specifically, PAAL mainly consists of an\nAccuracy Predictor (AP) and a Weighted Polling Strategy (WPS). The former is an\nattached learnable module that can accurately predict the segmentation accuracy\nof unlabeled samples relative to the target model with the predicted posterior\nprobability. The latter provides an efficient hybrid querying scheme by\ncombining predicted accuracy and feature representation, aiming to ensure the\nuncertainty and diversity of the acquired samples. Extensive experiment results\non multiple datasets demonstrate the superiority of PAAL. PAAL achieves\ncomparable accuracy to fully annotated data while reducing annotation costs by\napproximately 50% to 80%, showcasing significant potential in clinical\napplications. The code is available at https://github.com/shijun18/PAAL-MedSeg.\n", "date": "2024-05-01", "categories": []}
{"id": "2410.15605", "title": "Deep Active Learning with Manifold-preserving Trajectory Sampling", "abstract": "  Active learning (AL) is for optimizing the selection of unlabeled data for\nannotation (labeling), aiming to enhance model performance while minimizing\nlabeling effort. The key question in AL is which unlabeled data should be\nselected for annotation. Existing deep AL methods arguably suffer from bias\nincurred by clabeled data, which takes a much lower percentage than unlabeled\ndata in AL context. We observe that such an issue is severe in different types\nof data, such as vision and non-vision data. To address this issue, we propose\na novel method, namely Manifold-Preserving Trajectory Sampling (MPTS), aiming\nto enforce the feature space learned from labeled data to represent a more\naccurate manifold. By doing so, we expect to effectively correct the bias\nincurred by labeled data, which can cause a biased selection of unlabeled data.\nDespite its focus on manifold, the proposed method can be conveniently\nimplemented by performing distribution mapping with MMD (Maximum Mean\nDiscrepancies). Extensive experiments on various vision and non-vision\nbenchmark datasets demonstrate the superiority of our method. Our source code\ncan be found here.\n", "date": "2024-10-21", "categories": []}
{"id": "2312.05328", "title": "Bad Students Make Great Teachers: Active Learning Accelerates\n  Large-Scale Visual Understanding", "abstract": "  Power-law scaling indicates that large-scale training with uniform sampling\nis prohibitively slow. Active learning methods aim to increase data efficiency\nby prioritizing learning on the most relevant examples. Despite their appeal,\nthese methods have yet to be widely adopted since no one algorithm has been\nshown to a) generalize across models and tasks b) scale to large datasets and\nc) yield overall FLOP savings when accounting for the overhead of data\nselection. In this work we propose a method which satisfies these three\nproperties, leveraging small, cheap proxy models to estimate \"learnability\"\nscores for datapoints, which are used to prioritize data for the training of\nmuch larger models. As a result, our models require 46% and 51% fewer training\nupdates and up to 25% less total computation to reach the same performance as\nuniformly trained visual classifiers on JFT and multimodal models on ALIGN.\nFinally, we find our data-prioritization scheme to be complementary with recent\ndata-curation and learning objectives, yielding a new state-of-the-art in\nseveral multimodal transfer tasks.\n", "date": "2023-12-08", "categories": []}
{"id": "2404.05623", "title": "AnchorAL: Computationally Efficient Active Learning for Large and\n  Imbalanced Datasets", "abstract": "  Active learning for imbalanced classification tasks is challenging as the\nminority classes naturally occur rarely. Gathering a large pool of unlabelled\ndata is thus essential to capture minority instances. Standard pool-based\nactive learning is computationally expensive on large pools and often reaches\nlow accuracy by overfitting the initial decision boundary, thus failing to\nexplore the input space and find minority instances. To address these issues we\npropose AnchorAL. At each iteration, AnchorAL chooses class-specific instances\nfrom the labelled set, or anchors, and retrieves the most similar unlabelled\ninstances from the pool. This resulting subpool is then used for active\nlearning. Using a small, fixed-sized subpool AnchorAL allows scaling any active\nlearning strategy to large pools. By dynamically selecting different anchors at\neach iteration it promotes class balance and prevents overfitting the initial\ndecision boundary, thus promoting the discovery of new clusters of minority\ninstances. In experiments across different classification tasks, active\nlearning strategies, and model architectures AnchorAL is (i) faster, often\nreducing runtime from hours to minutes, (ii) trains more performant models,\n(iii) and returns more balanced datasets than competing methods.\n", "date": "2024-04-08", "categories": []}
{"id": "2410.11114", "title": "Active Learning for Robust and Representative LLM Generation in\n  Safety-Critical Scenarios", "abstract": "  Ensuring robust safety measures across a wide range of scenarios is crucial\nfor user-facing systems. While Large Language Models (LLMs) can generate\nvaluable data for safety measures, they often exhibit distributional biases,\nfocusing on common scenarios and neglecting rare but critical cases. This can\nundermine the effectiveness of safety protocols developed using such data. To\naddress this, we propose a novel framework that integrates active learning with\nclustering to guide LLM generation, enhancing their representativeness and\nrobustness in safety scenarios. We demonstrate the effectiveness of our\napproach by constructing a dataset of 5.4K potential safety violations through\nan iterative process involving LLM generation and an active learner model's\nfeedback. Our results show that the proposed framework produces a more\nrepresentative set of safety scenarios without requiring prior knowledge of the\nunderlying data distribution. Additionally, data acquired through our method\nimproves the accuracy and F1 score of both the active learner model as well\nmodels outside the scope of active learning process, highlighting its broad\napplicability.\n", "date": "2024-10-14", "categories": []}
{"id": "2306.05843", "title": "Adaptive Batch Sizes for Active Learning A Probabilistic Numerics\n  Approach", "abstract": "  Active learning parallelization is widely used, but typically relies on\nfixing the batch size throughout experimentation. This fixed approach is\ninefficient because of a dynamic trade-off between cost and speed -- larger\nbatches are more costly, smaller batches lead to slower wall-clock run-times --\nand the trade-off may change over the run (larger batches are often preferable\nearlier). To address this trade-off, we propose a novel Probabilistic Numerics\nframework that adaptively changes batch sizes. By framing batch selection as a\nquadrature task, our integration-error-aware algorithm facilitates the\nautomatic tuning of batch sizes to meet predefined quadrature precision\nobjectives, akin to how typical optimizers terminate based on convergence\nthresholds. This approach obviates the necessity for exhaustive searches across\nall potential batch sizes. We also extend this to scenarios with constrained\nactive learning and constrained optimization, interpreting constraint\nviolations as reductions in the precision requirement, to subsequently adapt\nbatch construction. Through extensive experiments, we demonstrate that our\napproach significantly enhances learning efficiency and flexibility in diverse\nBayesian batch active learning and Bayesian optimization applications.\n", "date": "2023-06-09", "categories": []}
{"id": "2410.08972", "title": "ALVIN: Active Learning Via INterpolation", "abstract": "  Active Learning aims to minimize annotation effort by selecting the most\nuseful instances from a pool of unlabeled data. However, typical active\nlearning methods overlook the presence of distinct example groups within a\nclass, whose prevalence may vary, e.g., in occupation classification datasets\ncertain demographics are disproportionately represented in specific classes.\nThis oversight causes models to rely on shortcuts for predictions, i.e.,\nspurious correlations between input attributes and labels occurring in\nwell-represented groups. To address this issue, we propose Active Learning Via\nINterpolation (ALVIN), which conducts intra-class interpolations between\nexamples from under-represented and well-represented groups to create anchors,\ni.e., artificial points situated between the example groups in the\nrepresentation space. By selecting instances close to the anchors for\nannotation, ALVIN identifies informative examples exposing the model to regions\nof the representation space that counteract the influence of shortcuts.\nCrucially, since the model considers these examples to be of high certainty,\nthey are likely to be ignored by typical active learning methods. Experimental\nresults on six datasets encompassing sentiment analysis, natural language\ninference, and paraphrase detection demonstrate that ALVIN outperforms\nstate-of-the-art active learning methods in both in-distribution and\nout-of-distribution generalization.\n", "date": "2024-10-11", "categories": []}
{"id": "2409.09258", "title": "Active Learning to Guide Labeling Efforts for Question Difficulty\n  Estimation", "abstract": "  In recent years, there has been a surge in research on Question Difficulty\nEstimation (QDE) using natural language processing techniques.\nTransformer-based neural networks achieve state-of-the-art performance,\nprimarily through supervised methods but with an isolated study in unsupervised\nlearning. While supervised methods focus on predictive performance, they\nrequire abundant labeled data. On the other hand, unsupervised methods do not\nrequire labeled data but rely on a different evaluation metric that is also\ncomputationally expensive in practice. This work bridges the research gap by\nexploring active learning for QDE, a supervised human-in-the-loop approach\nstriving to minimize the labeling efforts while matching the performance of\nstate-of-the-art models. The active learning process iteratively trains on a\nlabeled subset, acquiring labels from human experts only for the most\ninformative unlabeled data points. Furthermore, we propose a novel acquisition\nfunction PowerVariance to add the most informative samples to the labeled set,\na regression extension to the PowerBALD function popular in classification. We\nemploy DistilBERT for QDE and identify informative samples by applying Monte\nCarlo dropout to capture epistemic uncertainty in unlabeled samples. The\nexperiments demonstrate that active learning with PowerVariance acquisition\nachieves a performance close to fully supervised models after labeling only 10%\nof the training data. The proposed methodology promotes the responsible use of\neducational resources, makes QDE tools more accessible to course instructors,\nand is promising for other applications such as personalized support systems\nand question-answering tools.\n", "date": "2024-09-14", "categories": []}
{"id": "2410.07282", "title": "A Utility-Mining-Driven Active Learning Approach for Analyzing\n  Clickstream Sequences", "abstract": "  In rapidly evolving e-commerce industry, the capability of selecting\nhigh-quality data for model training is essential. This study introduces the\nHigh-Utility Sequential Pattern Mining using SHAP values (HUSPM-SHAP) model, a\nutility mining-based active learning strategy to tackle this challenge. We\nfound that the parameter settings for positive and negative SHAP values impact\nthe model's mining outcomes, introducing a key consideration into the active\nlearning framework. Through extensive experiments aimed at predicting behaviors\nthat do lead to purchases or not, the designed HUSPM-SHAP model demonstrates\nits superiority across diverse scenarios. The model's ability to mitigate\nlabeling needs while maintaining high predictive performance is highlighted.\nOur findings demonstrate the model's capability to refine e-commerce data\nprocessing, steering towards more streamlined, cost-effective prediction\nmodeling.\n", "date": "2024-10-09", "categories": []}
{"id": "2410.05660", "title": "Robust Transfer Learning for Active Level Set Estimation with Locally\n  Adaptive Gaussian Process Prior", "abstract": "  The objective of active level set estimation for a black-box function is to\nprecisely identify regions where the function values exceed or fall below a\nspecified threshold by iteratively performing function evaluations to gather\nmore information about the function. This becomes particularly important when\nfunction evaluations are costly, drastically limiting our ability to acquire\nlarge datasets. A promising way to sample-efficiently model the black-box\nfunction is by incorporating prior knowledge from a related function. However,\nthis approach risks slowing down the estimation task if the prior knowledge is\nirrelevant or misleading. In this paper, we present a novel transfer learning\nmethod for active level set estimation that safely integrates a given prior\nknowledge while constantly adjusting it to guarantee a robust performance of a\nlevel set estimation algorithm even when the prior knowledge is irrelevant. We\ntheoretically analyze this algorithm to show that it has a better level set\nconvergence compared to standard transfer learning approaches that do not make\nany adjustment to the prior. Additionally, extensive experiments across\nmultiple datasets confirm the effectiveness of our method when applied to\nvarious different level set estimation algorithms as well as different transfer\nlearning scenarios.\n", "date": "2024-10-08", "categories": []}
{"id": "2410.03915", "title": "Distribution Guided Active Feature Acquisition", "abstract": "  Human agents routinely reason on instances with incomplete and muddied data\n(and weigh the cost of obtaining further features). In contrast, much of ML is\ndevoted to the unrealistic, sterile environment where all features are observed\nand further information on an instance is obviated. Here we extend past static\nML and develop an active feature acquisition (AFA) framework that interacts\nwith the environment to obtain new information on-the-fly and can: 1) make\ninferences on an instance in the face of incomplete features, 2) determine a\nplan for feature acquisitions to obtain additional information on the instance\nat hand. We build our AFA framework on a backbone of understanding the\ninformation and conditional dependencies that are present in the data. First,\nwe show how to build generative models that can capture dependencies over\narbitrary subsets of features and employ these models for acquisitions in a\ngreedy scheme. After, we show that it is possible to guide the training of RL\nagents for AFA via side-information and auxiliary rewards stemming from our\ngenerative models. We also examine two important factors for deploying AFA\nmodels in real-world scenarios, namely interpretability and robustness.\nExtensive experiments demonstrate the state-of-the-art performance of our AFA\nframework.\n", "date": "2024-10-04", "categories": []}
{"id": "2410.04275", "title": "Language Model-Driven Data Pruning Enables Efficient Active Learning", "abstract": "  Active learning (AL) optimizes data labeling efficiency by selecting the most\ninformative instances for annotation. A key component in this procedure is an\nacquisition function that guides the selection process and identifies the\nsuitable instances for labeling from the unlabeled pool. However, these\nacquisition methods suffer from high computational costs with large unlabeled\ndata pools, posing a roadblock to their applicability on large datasets. To\naddress this challenge and bridge this gap, we introduce a novel plug-and-play\nunlabeled data pruning strategy, ActivePrune, which leverages language models\nto prune the unlabeled pool. ActivePrune implements a two-stage pruning\nprocess: an initial fast evaluation using perplexity scores from an n-gram\nlanguage model, followed by a high-quality selection using metrics for data\nquality computed through a quantized LLM. Additionally, to enhance the\ndiversity in the unlabeled pool, we propose a novel perplexity reweighting\nmethod that systematically brings forward underrepresented instances for\nselection in subsequent labeling iterations. Experiments on translation,\nsentiment analysis, topic classification, and summarization tasks on four\ndiverse datasets and four active learning strategies demonstrate that\nActivePrune outperforms existing data pruning methods. Finally, we compare the\nselection quality $\\leftrightarrow$ efficiency tradeoff of the data pruning\nmethods and demonstrate that ActivePrune is computationally more efficient than\nother LLM score-based pruning methods, and provides up to 74% reduction in the\nend-to-end time required for active learning.\n", "date": "2024-10-05", "categories": []}
{"id": "2410.04880", "title": "Improved detection of discarded fish species through BoxAL active\n  learning", "abstract": "  In recent years, powerful data-driven deep-learning techniques have been\ndeveloped and applied for automated catch registration. However, these methods\nare dependent on the labelled data, which is time-consuming, labour-intensive,\nexpensive to collect and need expert knowledge. In this study, we present an\nactive learning technique, named BoxAL, which includes estimation of epistemic\ncertainty of the Faster R-CNN object-detection model. The method allows\nselecting the most uncertain training images from an unlabeled pool, which are\nthen used to train the object-detection model. To evaluate the method, we used\nan open-source image dataset obtained with a dedicated image-acquisition system\ndeveloped for commercial trawlers targeting demersal species. We demonstrated,\nthat our approach allows reaching the same object-detection performance as with\nthe random sampling using 400 fewer labelled images. Besides, mean AP score was\nsignificantly higher at the last training iteration with 1100 training images,\nspecifically, 39.0&plusmn;1.6 and 34.8&plusmn;1.8 for certainty-based sampling\nand random sampling, respectively. Additionally, we showed that epistemic\ncertainty is a suitable method to sample images that the current iteration of\nthe model cannot deal with yet. Our study additionally showed that the sampled\nnew data is more valuable for training than the remaining unlabeled data. Our\nsoftware is available on https://github.com/pieterblok/boxal.\n", "date": "2024-10-07", "categories": []}
{"id": "2403.15744", "title": "On the Fragility of Active Learners for Text Classification", "abstract": "  Active learning (AL) techniques optimally utilize a labeling budget by\niteratively selecting instances that are most valuable for learning. However,\nthey lack ``prerequisite checks'', i.e., there are no prescribed criteria to\npick an AL algorithm best suited for a dataset. A practitioner must pick a\ntechnique they \\emph{trust} would beat random sampling, based on prior reported\nresults, and hope that it is resilient to the many variables in their\nenvironment: dataset, labeling budget and prediction pipelines. The important\nquestions then are: how often on average, do we expect any AL technique to\nreliably beat the computationally cheap and easy-to-implement strategy of\nrandom sampling? Does it at least make sense to use AL in an ``Always ON'' mode\nin a prediction pipeline, so that while it might not always help, it never\nunder-performs random sampling? How much of a role does the prediction pipeline\nplay in AL's success?\n  We examine these questions in detail for the task of text classification\nusing pre-trained representations, which are ubiquitous today.\n  Our primary contribution here is a rigorous evaluation of AL techniques, old\nand new, across setups that vary wrt datasets, text representations and\nclassifiers. This unlocks multiple insights around warm-up times, i.e., number\nof labels before gains from AL are seen, viability of an ``Always ON'' mode and\nthe relative significance of different factors. Additionally, we release a\nframework for rigorous benchmarking of AL techniques for text classification.\n", "date": "2024-03-23", "categories": []}
{"id": "2406.09206", "title": "Self-Training for Sample-Efficient Active Learning for Text\n  Classification with Pre-Trained Language Models", "abstract": "  Active learning is an iterative labeling process that is used to obtain a\nsmall labeled subset, despite the absence of labeled data, thereby enabling to\ntrain a model for supervised tasks such as text classification. While active\nlearning has made considerable progress in recent years due to improvements\nprovided by pre-trained language models, there is untapped potential in the\noften neglected unlabeled portion of the data, although it is available in\nconsiderably larger quantities than the usually small set of labeled data. In\nthis work, we investigate how self-training, a semi-supervised approach that\nuses a model to obtain pseudo-labels for unlabeled data, can be used to improve\nthe efficiency of active learning for text classification. Building on a\ncomprehensive reproduction of four previous self-training approaches, some of\nwhich are evaluated for the first time in the context of active learning or\nnatural language processing, we introduce HAST, a new and effective\nself-training strategy, which is evaluated on four text classification\nbenchmarks. Our results show that it outperforms the reproduced self-training\napproaches and reaches classification results comparable to previous\nexperiments for three out of four datasets, using as little as 25% of the data.\nThe code is publicly available at\nhttps://github.com/chschroeder/self-training-for-sample-efficient-active-learning .\n", "date": "2024-06-13", "categories": []}
{"id": "2401.05467", "title": "Can Active Label Correction Improve LLM-based Modular AI Systems?", "abstract": "  Modular AI systems can be developed using LLM-prompts-based modules to\nminimize deployment time even for complex tasks. However, these systems do not\nalways perform well and improving them using the data traces collected from a\ndeployment remains an open challenge. The data traces contain LLM inputs and\noutputs, but the annotations from LLMs are noisy. We hypothesize that Active\nLabel Correction (ALC) can be use on the collected data to train smaller\ntask-specific improved models that can replace LLM-based modules. In this\npaper, we study the noise in three GPT-3.5-annotated datasets and their\ndenoising with human feedback. We also propose a novel method ALC3 that\niteratively applies three updates to the training dataset: auto-correction,\ncorrection using human feedback and filtering. Our results show that ALC3 can\nlead to oracle performance with feedback on 17-24% fewer examples than the\nnumber of noisy examples in the dataset across three different NLP tasks.\n", "date": "2024-01-10", "categories": []}
{"id": "2405.14121", "title": "One-shot Active Learning Based on Lewis Weight Sampling for Multiple\n  Deep Models", "abstract": "  Active learning (AL) for multiple target models aims to reduce labeled data\nquerying while effectively training multiple models concurrently. Existing AL\nalgorithms often rely on iterative model training, which can be computationally\nexpensive, particularly for deep models. In this paper, we propose a one-shot\nAL method to address this challenge, which performs all label queries without\nrepeated model training.\n  Specifically, we extract different representations of the same dataset using\ndistinct network backbones, and actively learn the linear prediction layer on\neach representation via an $\\ell_p$-regression formulation. The regression\nproblems are solved approximately by sampling and reweighting the unlabeled\ninstances based on their maximum Lewis weights across the representations. An\nupper bound on the number of samples needed is provided with a rigorous\nanalysis for $p\\in [1, +\\infty)$.\n  Experimental results on 11 benchmarks show that our one-shot approach\nachieves competitive performances with the state-of-the-art AL methods for\nmultiple target models.\n", "date": "2024-05-23", "categories": []}
{"id": "2405.17627", "title": "Salutary Labeling with Zero Human Annotation", "abstract": "  Active learning strategically selects informative unlabeled data points and\nqueries their ground truth labels for model training. The prevailing assumption\nunderlying this machine learning paradigm is that acquiring these ground truth\nlabels will optimally enhance model performance. However, this assumption may\nnot always hold true or maximize learning capacity, particularly considering\nthe costly labor annotations required for ground truth labels. In contrast to\ntraditional ground truth labeling, this paper proposes salutary labeling, which\nautomatically assigns the most beneficial labels to the most informative\nsamples without human annotation. Specifically, we utilize the influence\nfunction, a tool for estimating sample influence, to select newly added samples\nand assign their salutary labels by choosing the category that maximizes their\npositive influence. This process eliminates the need for human annotation.\nExtensive experiments conducted on nine benchmark datasets demonstrate the\nsuperior performance of our salutary labeling approach over traditional active\nlearning strategies. Additionally, we provide several in-depth explorations and\npractical applications of large language model (LLM) fine-tuning.\n", "date": "2024-05-27", "categories": []}
{"id": "2409.18561", "title": "AL-GTD: Deep Active Learning for Gaze Target Detection", "abstract": "  Gaze target detection aims at determining the image location where a person\nis looking. While existing studies have made significant progress in this area\nby regressing accurate gaze heatmaps, these achievements have largely relied on\naccess to extensive labeled datasets, which demands substantial human labor. In\nthis paper, our goal is to reduce the reliance on the size of labeled training\ndata for gaze target detection. To achieve this, we propose AL-GTD, an\ninnovative approach that integrates supervised and self-supervised losses\nwithin a novel sample acquisition function to perform active learning (AL).\nAdditionally, it utilizes pseudo-labeling to mitigate distribution shifts\nduring the training phase. AL-GTD achieves the best of all AUC results by\nutilizing only 40-50% of the training data, in contrast to state-of-the-art\n(SOTA) gaze target detectors requiring the entire training dataset to achieve\nthe same performance. Importantly, AL-GTD quickly reaches satisfactory\nperformance with 10-20% of the training data, showing the effectiveness of our\nacquisition function, which is able to acquire the most informative samples. We\nprovide a comprehensive experimental analysis by adapting several AL methods\nfor the task. AL-GTD outperforms AL competitors, simultaneously exhibiting\nsuperior performance compared to SOTA gaze target detectors when all are\ntrained within a low-data regime. Code is available at\nhttps://github.com/francescotonini/al-gtd.\n", "date": "2024-09-27", "categories": []}
{"id": "2409.18572", "title": "Towards an active-learning approach to resource allocation for\n  population-based damage prognosis", "abstract": "  Damage prognosis is, arguably, one of the most difficult tasks of structural\nhealth monitoring (SHM). To address common problems of damage prognosis, a\npopulation-based SHM (PBSHM) approach is adopted in the current work. In this\napproach the prognosis problem is considered as an information-sharing problem\nwhere data from past structures are exploited to make more accurate inferences\nregarding currently-degrading structures. For a given population, there may\nexist restrictions on the resources available to conduct monitoring; thus, the\ncurrent work studies the problem of allocating such resources within a\npopulation of degrading structures with a view to maximising the\ndamage-prognosis accuracy. The challenges of the current framework are mainly\nassociated with the inference of outliers on the level of damage evolution,\ngiven partial data from the damage-evolution phenomenon. The current approach\nconsiders an initial population of structures for which damage evolution is\nextensively observed. Subsequently, a second population of structures with\nevolving damage is considered for which two monitoring systems are available, a\nlow-availability and high-fidelity (low-uncertainty) one, and a\nwidely-available and low-fidelity (high-uncertainty) one. The task of the\ncurrent work is to follow an active-learning approach to identify the\nstructures to which the high-fidelity system should be assigned in order to\nenhance the predictive capabilities of the machine-learning model throughout\nthe population.\n", "date": "2024-09-27", "categories": []}
{"id": "2409.17607", "title": "Dirichlet-Based Coarse-to-Fine Example Selection For Open-Set Annotation", "abstract": "  Active learning (AL) has achieved great success by selecting the most\nvaluable examples from unlabeled data. However, they usually deteriorate in\nreal scenarios where open-set noise gets involved, which is studied as open-set\nannotation (OSA). In this paper, we owe the deterioration to the unreliable\npredictions arising from softmax-based translation invariance and propose a\nDirichlet-based Coarse-to-Fine Example Selection (DCFS) strategy accordingly.\nOur method introduces simplex-based evidential deep learning (EDL) to break\ntranslation invariance and distinguish known and unknown classes by considering\nevidence-based data and distribution uncertainty simultaneously. Furthermore,\nhard known-class examples are identified by model discrepancy generated from\ntwo classifier heads, where we amplify and alleviate the model discrepancy\nrespectively for unknown and known classes. Finally, we combine the discrepancy\nwith uncertainties to form a two-stage strategy, selecting the most informative\nexamples from known classes. Extensive experiments on various openness ratio\ndatasets demonstrate that DCFS achieves state-of-art performance.\n", "date": "2024-09-26", "categories": []}
{"id": "2409.18104", "title": "Find Rhinos without Finding Rhinos: Active Learning with Multimodal\n  Imagery of South African Rhino Habitats", "abstract": "  Much of Earth's charismatic megafauna is endangered by human activities,\nparticularly the rhino, which is at risk of extinction due to the poaching\ncrisis in Africa. Monitoring rhinos' movement is crucial to their protection\nbut has unfortunately proven difficult because rhinos are elusive. Therefore,\ninstead of tracking rhinos, we propose the novel approach of mapping communal\ndefecation sites, called middens, which give information about rhinos' spatial\nbehavior valuable to anti-poaching, management, and reintroduction efforts.\nThis paper provides the first-ever mapping of rhino midden locations by\nbuilding classifiers to detect them using remotely sensed thermal, RGB, and\nLiDAR imagery in passive and active learning settings. As existing active\nlearning methods perform poorly due to the extreme class imbalance in our\ndataset, we design MultimodAL, an active learning system employing a ranking\ntechnique and multimodality to achieve competitive performance with passive\nlearning models with 94% fewer labels. Our methods could therefore save over 76\nhours in labeling time when used on a similarly-sized dataset. Unexpectedly,\nour midden map reveals that rhino middens are not randomly distributed\nthroughout the landscape; rather, they are clustered. Consequently, rangers\nshould be targeted at areas with high midden densities to strengthen\nanti-poaching efforts, in line with UN Target 15.7.\n", "date": "2024-09-26", "categories": []}
{"id": "2405.12081", "title": "Selective Annotation via Data Allocation: These Data Should Be Triaged\n  to Experts for Annotation Rather Than the Model", "abstract": "  To obtain high-quality annotations under limited budget, semi-automatic\nannotation methods are commonly used, where a portion of the data is annotated\nby experts and a model is then trained to complete the annotations for the\nremaining data. However, these methods mainly focus on selecting informative\ndata for expert annotations to improve the model predictive ability (i.e.,\ntriage-to-human data), while the rest of the data is indiscriminately assigned\nto model annotation (i.e., triage-to-model data). This may lead to\ninefficiencies in budget allocation for annotations, as easy data that the\nmodel could accurately annotate may be unnecessarily assigned to the expert,\nand hard data may be misclassified by the model. As a result, the overall\nannotation quality may be compromised. To address this issue, we propose a\nselective annotation framework called SANT. It effectively takes advantage of\nboth the triage-to-human and triage-to-model data through the proposed\nerror-aware triage and bi-weighting mechanisms. As such, informative or hard\ndata is assigned to the expert for annotation, while easy data is handled by\nthe model. Experimental results show that SANT consistently outperforms other\nbaselines, leading to higher-quality annotation through its proper allocation\nof data to both expert and model workers. We provide pioneering work on data\nannotation within budget constraints, establishing a landmark for future\ntriage-based annotation studies.\n", "date": "2024-05-20", "categories": []}
{"id": "2409.11808", "title": "Accelerating the Training and Improving the Reliability of\n  Machine-Learned Interatomic Potentials for Strongly Anharmonic Materials\n  through Active Learning", "abstract": "  Molecular dynamics (MD) employing machine-learned interatomic potentials\n(MLIPs) serve as an efficient, urgently needed complement to ab initio\nmolecular dynamics (aiMD). By training these potentials on data generated from\nab initio methods, their averaged predictions can exhibit comparable\nperformance to ab initio methods at a fraction of the cost. However,\ninsufficient training sets might lead to an improper description of the\ndynamics in strongly anharmonic materials, because critical effects might be\noverlooked in relevant cases, or only incorrectly captured, or hallucinated by\nthe MLIP when they are not actually present. In this work, we show that an\nactive learning scheme that combines MD with MLIPs (MLIP-MD) and uncertainty\nestimates can avoid such problematic predictions. In short, efficient MLIP-MD\nis used to explore configuration space quickly, whereby an acquisition function\nbased on uncertainty estimates and on energetic viability is employed to\nmaximize the value of the newly generated data and to focus on the most\nunfamiliar but reasonably accessible regions of phase space. To verify our\nmethodology, we screen over 112 materials and identify 10 examples experiencing\nthe aforementioned problems. Using CuI and AgGaSe$_2$ as archetypes for these\nproblematic materials, we discuss the physical implications for strongly\nanharmonic effects and demonstrate how the developed active learning scheme can\naddress these issues.\n", "date": "2024-09-18", "categories": []}
{"id": "2409.09078", "title": "Bounds on the Generalization Error in Active Learning", "abstract": "  We establish empirical risk minimization principles for active learning by\nderiving a family of upper bounds on the generalization error. Aligning with\nempirical observations, the bounds suggest that superior query algorithms can\nbe obtained by combining both informativeness and representativeness query\nstrategies, where the latter is assessed using integral probability metrics. To\nfacilitate the use of these bounds in application, we systematically link\ndiverse active learning scenarios, characterized by their loss functions and\nhypothesis classes to their corresponding upper bounds. Our results show that\nregularization techniques used to constraint the complexity of various\nhypothesis classes are sufficient conditions to ensure the validity of the\nbounds. The present work enables principled construction and empirical\nquality-evaluation of query algorithms in active learning.\n", "date": "2024-09-10", "categories": []}
{"id": "2409.05425", "title": "Distribution Discrepancy and Feature Heterogeneity for Active 3D Object\n  Detection", "abstract": "  LiDAR-based 3D object detection is a critical technology for the development\nof autonomous driving and robotics. However, the high cost of data annotation\nlimits its advancement. We propose a novel and effective active learning (AL)\nmethod called Distribution Discrepancy and Feature Heterogeneity (DDFH), which\nsimultaneously considers geometric features and model embeddings, assessing\ninformation from both the instance-level and frame-level perspectives.\nDistribution Discrepancy evaluates the difference and novelty of instances\nwithin the unlabeled and labeled distributions, enabling the model to learn\nefficiently with limited data. Feature Heterogeneity ensures the heterogeneity\nof intra-frame instance features, maintaining feature diversity while avoiding\nredundant or similar instances, thus minimizing annotation costs. Finally,\nmultiple indicators are efficiently aggregated using Quantile Transform,\nproviding a unified measure of informativeness. Extensive experiments\ndemonstrate that DDFH outperforms the current state-of-the-art (SOTA) methods\non the KITTI and Waymo datasets, effectively reducing the bounding box\nannotation cost by 56.3% and showing robustness when working with both\none-stage and two-stage models.\n", "date": "2024-09-09", "categories": []}
{"id": "2409.07379", "title": "FIRAL: An Active Learning Algorithm for Multinomial Logistic Regression", "abstract": "  We investigate theory and algorithms for pool-based active learning for\nmulticlass classification using multinomial logistic regression. Using finite\nsample analysis, we prove that the Fisher Information Ratio (FIR) lower and\nupper bounds the excess risk. Based on our theoretical analysis, we propose an\nactive learning algorithm that employs regret minimization to minimize the FIR.\nTo verify our derived excess risk bounds, we conduct experiments on synthetic\ndatasets. Furthermore, we compare FIRAL with five other methods and found that\nour scheme outperforms them: it consistently produces the smallest\nclassification error in the multiclass logistic regression setting, as\ndemonstrated through experiments on MNIST, CIFAR-10, and 50-class ImageNet.\n", "date": "2024-09-11", "categories": []}
{"id": "2409.07392", "title": "A Scalable Algorithm for Active Learning", "abstract": "  FIRAL is a recently proposed deterministic active learning algorithm for\nmulticlass classification using logistic regression. It was shown to outperform\nthe state-of-the-art in terms of accuracy and robustness and comes with\ntheoretical performance guarantees. However, its scalability suffers when\ndealing with datasets featuring a large number of points $n$, dimensions $d$,\nand classes $c$, due to its $\\mathcal{O}(c^2d^2+nc^2d)$ storage and\n$\\mathcal{O}(c^3(nd^2 + bd^3 + bn))$ computational complexity where $b$ is the\nnumber of points to select in active learning. To address these challenges, we\npropose an approximate algorithm with storage requirements reduced to\n$\\mathcal{O}(n(d+c) + cd^2)$ and a computational complexity of\n$\\mathcal{O}(bncd^2)$. Additionally, we present a parallel implementation on\nGPUs. We demonstrate the accuracy and scalability of our approach using MNIST,\nCIFAR-10, Caltech101, and ImageNet. The accuracy tests reveal no deterioration\nin accuracy compared to FIRAL. We report strong and weak scaling tests on up to\n12 GPUs, for three million point synthetic dataset.\n", "date": "2024-09-11", "categories": []}
{"id": "2306.11715", "title": "Multi-Fidelity Active Learning with GFlowNets", "abstract": "  In the last decades, the capacity to generate large amounts of data in\nscience and engineering applications has been growing steadily. Meanwhile,\nmachine learning has progressed to become a suitable tool to process and\nutilise the available data. Nonetheless, many relevant scientific and\nengineering problems present challenges where current machine learning methods\ncannot yet efficiently leverage the available data and resources. For example,\nin scientific discovery, we are often faced with the problem of exploring very\nlarge, structured and high-dimensional spaces. Moreover, the high fidelity,\nblack-box objective function is often very expensive to evaluate. Progress in\nmachine learning methods that can efficiently tackle such challenges would help\naccelerate currently crucial areas such as drug and materials discovery. In\nthis paper, we propose a multi-fidelity active learning algorithm with\nGFlowNets as a sampler, to efficiently discover diverse, high-scoring\ncandidates where multiple approximations of the black-box function are\navailable at lower fidelity and cost. Our evaluation on molecular discovery\ntasks shows that multi-fidelity active learning with GFlowNets can discover\nhigh-scoring candidates at a fraction of the budget of its single-fidelity\ncounterpart while maintaining diversity, unlike RL-based alternatives. These\nresults open new avenues for multi-fidelity active learning to accelerate\nscientific discovery and engineering design.\n", "date": "2023-06-20", "categories": []}
{"id": "2402.02198", "title": "Co-orchestration of Multiple Instruments to Uncover Structure-Property\n  Relationships in Combinatorial Libraries", "abstract": "  The rapid growth of automated and autonomous instrumentations brings forth an\nopportunity for the co-orchestration of multimodal tools, equipped with\nmultiple sequential detection methods, or several characterization tools to\nexplore identical samples. This can be exemplified by the combinatorial\nlibraries that can be explored in multiple locations by multiple tools\nsimultaneously, or downstream characterization in automated synthesis systems.\nIn the co-orchestration approaches, information gained in one modality should\naccelerate the discovery of other modalities. Correspondingly, the\norchestrating agent should select the measurement modality based on the\nanticipated knowledge gain and measurement cost. Here, we propose and implement\na co-orchestration approach for conducting measurements with complex\nobservables such as spectra or images. The method relies on combining\ndimensionality reduction by variational autoencoders with representation\nlearning for control over the latent space structure, and integrated into\niterative workflow via multi-task Gaussian Processes (GP). This approach\nfurther allows for the native incorporation of the system's physics via a\nprobabilistic model as a mean function of the GP. We illustrated this method\nfor different modalities of piezoresponse force microscopy and micro-Raman on\ncombinatorial $Sm-BiFeO_3$ library. However, the proposed framework is general\nand can be extended to multiple measurement modalities and arbitrary\ndimensionality of measured signals. The analysis code that supports the funding\nis publicly available at https://github.com/Slautin/2024_Co-orchestration.\n", "date": "2024-02-03", "categories": []}
{"id": "2404.12899", "title": "Bayesian Co-navigation: Dynamic Designing of the Materials Digital Twins\n  via Active Learning", "abstract": "  Scientific advancement is universally based on the dynamic interplay between\ntheoretical insights, modelling, and experimental discoveries. However, this\nfeedback loop is often slow, including delayed community interactions and the\ngradual integration of experimental data into theoretical frameworks. This\nchallenge is particularly exacerbated in domains dealing with high-dimensional\nobject spaces, such as molecules and complex microstructures. Hence, the\nintegration of theory within automated and autonomous experimental setups, or\ntheory in the loop automated experiment, is emerging as a crucial objective for\naccelerating scientific research. The critical aspect is not only to use theory\nbut also on-the-fly theory updates during the experiment. Here, we introduce a\nmethod for integrating theory into the loop through Bayesian co-navigation of\ntheoretical model space and experimentation. Our approach leverages the\nconcurrent development of surrogate models for both simulation and experimental\ndomains at the rates determined by latencies and costs of experiments and\ncomputation, alongside the adjustment of control parameters within theoretical\nmodels to minimize epistemic uncertainty over the experimental object spaces.\nThis methodology facilitates the creation of digital twins of material\nstructures, encompassing both the surrogate model of behavior that includes the\ncorrelative part and the theoretical model itself. While demonstrated here\nwithin the context of functional responses in ferroelectric materials, our\napproach holds promise for broader applications, the exploration of optical\nproperties in nanoclusters, microstructure-dependent properties in complex\nmaterials, and properties of molecular systems. The analysis code that supports\nthe funding is publicly available at\nhttps://github.com/Slautin/2024_Co-navigation/tree/main\n", "date": "2024-04-19", "categories": []}
{"id": "2408.13690", "title": "Understanding Uncertainty-based Active Learning Under Model Mismatch", "abstract": "  Instead of randomly acquiring training data points, Uncertainty-based Active\nLearning (UAL) operates by querying the label(s) of pivotal samples from an\nunlabeled pool selected based on the prediction uncertainty, thereby aiming at\nminimizing the labeling cost for model training. The efficacy of UAL critically\ndepends on the model capacity as well as the adopted uncertainty-based\nacquisition function. Within the context of this study, our analytical focus is\ndirected toward comprehending how the capacity of the machine learning model\nmay affect UAL efficacy. Through theoretical analysis, comprehensive\nsimulations, and empirical studies, we conclusively demonstrate that UAL can\nlead to worse performance in comparison with random sampling when the machine\nlearning model class has low capacity and is unable to cover the underlying\nground truth. In such situations, adopting acquisition functions that directly\ntarget estimating the prediction performance may be beneficial for improving\nthe performance of UAL.\n", "date": "2024-08-24", "categories": []}
{"id": "2408.12774", "title": "Semi-Supervised Variational Adversarial Active Learning via Learning to\n  Rank and Agreement-Based Pseudo Labeling", "abstract": "  Active learning aims to alleviate the amount of labor involved in data\nlabeling by automating the selection of unlabeled samples via an acquisition\nfunction. For example, variational adversarial active learning (VAAL) leverages\nan adversarial network to discriminate unlabeled samples from labeled ones\nusing latent space information. However, VAAL has the following shortcomings:\n(i) it does not exploit target task information, and (ii) unlabeled data is\nonly used for sample selection rather than model training. To address these\nlimitations, we introduce novel techniques that significantly improve the use\nof abundant unlabeled data during training and take into account the task\ninformation. Concretely, we propose an improved pseudo-labeling algorithm that\nleverages information from all unlabeled data in a semi-supervised manner, thus\nallowing a model to explore a richer data space. In addition, we develop a\nranking-based loss prediction module that converts predicted relative ranking\ninformation into a differentiable ranking loss. This loss can be embedded as a\nrank variable into the latent space of a variational autoencoder and then\ntrained with a discriminator in an adversarial fashion for sample selection. We\ndemonstrate the superior performance of our approach over the state of the art\non various image classification and segmentation benchmark datasets.\n", "date": "2024-08-23", "categories": []}
{"id": "2407.03895", "title": "Scoping Review of Active Learning Strategies and their Evaluation\n  Environments for Entity Recognition Tasks", "abstract": "  We conducted a scoping review for active learning in the domain of natural\nlanguage processing (NLP), which we summarize in accordance with the PRISMA-ScR\nguidelines as follows:\n  Objective: Identify active learning strategies that were proposed for entity\nrecognition and their evaluation environments (datasets, metrics, hardware,\nexecution time). Design: We used Scopus and ACM as our search engines. We\ncompared the results with two literature surveys to assess the search quality.\nWe included peer-reviewed English publications introducing or comparing active\nlearning strategies for entity recognition. Results: We analyzed 62 relevant\npapers and identified 106 active learning strategies. We grouped them into\nthree categories: exploitation-based (60x), exploration-based (14x), and hybrid\nstrategies (32x). We found that all studies used the F1-score as an evaluation\nmetric. Information about hardware (6x) and execution time (13x) was only\noccasionally included. The 62 papers used 57 different datasets to evaluate\ntheir respective strategies. Most datasets contained newspaper articles or\nbiomedical/medical data. Our analysis revealed that 26 out of 57 datasets are\npublicly accessible.\n  Conclusion: Numerous active learning strategies have been identified, along\nwith significant open questions that still need to be addressed. Researchers\nand practitioners face difficulties when making data-driven decisions about\nwhich active learning strategy to adopt. Conducting comprehensive empirical\ncomparisons using the evaluation environment proposed in this study could help\nestablish best practices in the domain.\n", "date": "2024-07-04", "categories": []}
{"id": "2408.11191", "title": "Active Learning of Molecular Data for Task-Specific Objectives", "abstract": "  Active learning (AL) has shown promise for being a particularly\ndata-efficient machine learning approach. Yet, its performance depends on the\napplication and it is not clear when AL practitioners can expect computational\nsavings. Here, we carry out a systematic AL performance assessment for three\ndiverse molecular datasets and two common scientific tasks: compiling compact,\ninformative datasets and targeted molecular searches. We implemented AL with\nGaussian processes (GP) and used the many-body tensor as molecular\nrepresentation. For the first task, we tested different data acquisition\nstrategies, batch sizes and GP noise settings. AL was insensitive to the\nacquisition batch size and we observed the best AL performance for the\nacquisition strategy that combines uncertainty reduction with clustering to\npromote diversity. However, for optimal GP noise settings, AL did not\noutperform randomized selection of data points. Conversely, for targeted\nsearches, AL outperformed random sampling and achieved data savings up to 64%.\nOur analysis provides insight into this task-specific performance difference in\nterms of target distributions and data collection strategies. We established\nthat the performance of AL depends on the relative distribution of the target\nmolecules in comparison to the total dataset distribution, with the largest\ncomputational savings achieved when their overlap is minimal.\n", "date": "2024-08-20", "categories": []}
{"id": "2408.09914", "title": "Active Learning for Identifying Disaster-Related Tweets: A Comparison\n  with Keyword Filtering and Generic Fine-Tuning", "abstract": "  Information from social media can provide essential information for emergency\nresponse during natural disasters in near real-time. However, it is difficult\nto identify the disaster-related posts among the large amounts of unstructured\ndata available. Previous methods often use keyword filtering, topic modelling\nor classification-based techniques to identify such posts. Active Learning (AL)\npresents a promising sub-field of Machine Learning (ML) that has not been used\nmuch in the field of text classification of social media content. This study\ntherefore investigates the potential of AL for identifying disaster-related\nTweets. We compare a keyword filtering approach, a RoBERTa model fine-tuned\nwith generic data from CrisisLex, a base RoBERTa model trained with AL and a\nfine-tuned RoBERTa model trained with AL regarding classification performance.\nFor testing, data from CrisisLex and manually labelled data from the 2021 flood\nin Germany and the 2023 Chile forest fires were considered. The results show\nthat generic fine-tuning combined with 10 rounds of AL outperformed all other\napproaches. Consequently, a broadly applicable model for the identification of\ndisaster-related Tweets could be trained with very little labelling effort. The\nmodel can be applied to use cases beyond this study and provides a useful tool\nfor further research in social media analysis.\n", "date": "2024-08-19", "categories": []}
{"id": "2307.10988", "title": "On minimizing the training set fill distance in machine learning\n  regression", "abstract": "  For regression tasks one often leverages large datasets for training\npredictive machine learning models. However, using large datasets may not be\nfeasible due to computational limitations or high data labelling costs.\nTherefore, suitably selecting small training sets from large pools of\nunlabelled data points is essential to maximize model performance while\nmaintaining efficiency. In this work, we study Farthest Point Sampling (FPS), a\ndata selection approach that aims to minimize the fill distance of the selected\nset. We derive an upper bound for the maximum expected prediction error,\nconditional to the location of the unlabelled data points, that linearly\ndepends on the training set fill distance. For empirical validation, we perform\nexperiments using two regression models on three datasets. We empirically show\nthat selecting a training set by aiming to minimize the fill distance, thereby\nminimizing our derived bound, significantly reduces the maximum prediction\nerror of various regression models, outperforming alternative sampling\napproaches by a large margin. Furthermore, we show that selecting training sets\nwith the FPS can also increase model stability for the specific case of\nGaussian kernel regression approaches.\n", "date": "2023-07-20", "categories": []}
{"id": "2404.09030", "title": "Active Learning for Control-Oriented Identification of Nonlinear Systems", "abstract": "  Model-based reinforcement learning is an effective approach for controlling\nan unknown system. It is based on a longstanding pipeline familiar to the\ncontrol community in which one performs experiments on the environment to\ncollect a dataset, uses the resulting dataset to identify a model of the\nsystem, and finally performs control synthesis using the identified model. As\ninteracting with the system may be costly and time consuming, targeted\nexploration is crucial for developing an effective control-oriented model with\nminimal experimentation. Motivated by this challenge, recent work has begun to\nstudy finite sample data requirements and sample efficient algorithms for the\nproblem of optimal exploration in model-based reinforcement learning. However,\nexisting theory and algorithms are limited to model classes which are linear in\nthe parameters. Our work instead focuses on models with nonlinear parameter\ndependencies, and presents the first finite sample analysis of an active\nlearning algorithm suitable for a general class of nonlinear dynamics. In\ncertain settings, the excess control cost of our algorithm achieves the optimal\nrate, up to logarithmic factors. We validate our approach in simulation,\nshowcasing the advantage of active, control-oriented exploration for\ncontrolling nonlinear systems.\n", "date": "2024-04-13", "categories": []}
{"id": "2408.04482", "title": "SegXAL: Explainable Active Learning for Semantic Segmentation in Driving\n  Scene Scenarios", "abstract": "  Most of the sophisticated AI models utilize huge amounts of annotated data\nand heavy training to achieve high-end performance. However, there are certain\nchallenges that hinder the deployment of AI models \"in-the-wild\" scenarios,\ni.e., inefficient use of unlabeled data, lack of incorporation of human\nexpertise, and lack of interpretation of the results. To mitigate these\nchallenges, we propose a novel Explainable Active Learning (XAL) model,\nXAL-based semantic segmentation model \"SegXAL\", that can (i) effectively\nutilize the unlabeled data, (ii) facilitate the \"Human-in-the-loop\" paradigm,\nand (iii) augment the model decisions in an interpretable way. In particular,\nwe investigate the application of the SegXAL model for semantic segmentation in\ndriving scene scenarios. The SegXAL model proposes the image regions that\nrequire labeling assistance from Oracle by dint of explainable AI (XAI) and\nuncertainty measures in a weakly-supervised manner. Specifically, we propose a\nnovel Proximity-aware Explainable-AI (PAE) module and Entropy-based Uncertainty\n(EBU) module to get an Explainable Error Mask, which enables the machine\nteachers/human experts to provide intuitive reasoning behind the results and to\nsolicit feedback to the AI system via an active learning strategy. Such a\nmechanism bridges the semantic gap between man and machine through\ncollaborative intelligence, where humans and AI actively enhance each other's\ncomplementary strengths. A novel high-confidence sample selection technique\nbased on the DICE similarity coefficient is also presented within the SegXAL\nframework. Extensive quantitative and qualitative analyses are carried out in\nthe benchmarking Cityscape dataset. Results show the outperformance of our\nproposed SegXAL against other state-of-the-art models.\n", "date": "2024-08-08", "categories": []}
{"id": "2408.02849", "title": "Active Learning for WBAN-based Health Monitoring", "abstract": "  We consider a novel active learning problem motivated by the need of learning\nmachine learning models for health monitoring in wireless body area network\n(WBAN). Due to the limited resources at body sensors, collecting each unlabeled\nsample in WBAN incurs a nontrivial cost. Moreover, training health monitoring\nmodels typically requires labels indicating the patient's health state that\nneed to be generated by healthcare professionals, which cannot be obtained at\nthe same pace as data collection. These challenges make our problem\nfundamentally different from classical active learning, where unlabeled samples\nare free and labels can be queried in real time. To handle these challenges, we\npropose a two-phased active learning method, consisting of an online phase\nwhere a coreset construction algorithm is proposed to select a subset of\nunlabeled samples based on their noisy predictions, and an offline phase where\nthe selected samples are labeled to train the target model. The samples\nselected by our algorithm are proved to yield a guaranteed error in\napproximating the full dataset in evaluating the loss function. Our evaluation\nbased on real health monitoring data and our own experimentation demonstrates\nthat our solution can drastically save the data curation cost without\nsacrificing the quality of the target model.\n", "date": "2024-08-05", "categories": []}
{"id": "2406.10107", "title": "Annotation Cost-Efficient Active Learning for Deep Metric Learning\n  Driven Remote Sensing Image Retrieval", "abstract": "  Deep metric learning (DML) has shown to be effective for content-based image\nretrieval (CBIR) in remote sensing (RS). Most of DML methods for CBIR rely on a\nhigh number of annotated images to accurately learn model parameters of deep\nneural networks (DNNs). However, gathering such data is time-consuming and\ncostly. To address this, we propose an annotation cost-efficient active\nlearning (ANNEAL) method tailored to DML-driven CBIR in RS. ANNEAL aims to\ncreate a small but informative training set made up of similar and dissimilar\nimage pairs to be utilized for accurately learning a metric space. The\ninformativeness of image pairs is evaluated by combining uncertainty and\ndiversity criteria. To assess the uncertainty of image pairs, we introduce two\nalgorithms: 1) metric-guided uncertainty estimation (MGUE); and 2) binary\nclassifier guided uncertainty estimation (BCGUE). MGUE algorithm automatically\nestimates a threshold value that acts as a boundary between similar and\ndissimilar image pairs based on the distances in the metric space. The closer\nthe similarity between image pairs is to the estimated threshold value the\nhigher their uncertainty. BCGUE algorithm estimates the uncertainty of the\nimage pairs based on the confidence of the classifier in assigning correct\nsimilarity labels. The diversity criterion is assessed through a\nclustering-based strategy. ANNEAL combines either MGUE or BCGUE algorithm with\nthe clustering-based strategy to select the most informative image pairs, which\nare then labelled by expert annotators as similar or dissimilar. This way of\nannotating images significantly reduces the annotation cost compared to\nannotating images with land-use land-cover class labels. Experimental results\non two RS benchmark datasets demonstrate the effectiveness of our method. The\ncode of this work is publicly available at\n\\url{https://git.tu-berlin.de/rsim/anneal_tgrs}.\n", "date": "2024-06-14", "categories": []}
{"id": "2408.01694", "title": "Bayesian Active Learning for Semantic Segmentation", "abstract": "  Fully supervised training of semantic segmentation models is costly and\nchallenging because each pixel within an image needs to be labeled. Therefore,\nthe sparse pixel-level annotation methods have been introduced to train models\nwith a subset of pixels within each image. We introduce a Bayesian active\nlearning framework based on sparse pixel-level annotation that utilizes a\npixel-level Bayesian uncertainty measure based on Balanced Entropy (BalEnt)\n[84]. BalEnt captures the information between the models' predicted\nmarginalized probability distribution and the pixel labels. BalEnt has linear\nscalability with a closed analytical form and can be calculated independently\nper pixel without relational computations with other pixels. We train our\nproposed active learning framework for Cityscapes, Camvid, ADE20K and VOC2012\nbenchmark datasets and show that it reaches supervised levels of mIoU using\nonly a fraction of labeled pixels while outperforming the previous\nstate-of-the-art active learning models with a large margin.\n", "date": "2024-08-03", "categories": []}
{"id": "2408.01861", "title": "Batch Active Learning in Gaussian Process Regression using Derivatives", "abstract": "  We investigate the use of derivative information for Batch Active Learning in\nGaussian Process regression models. The proposed approach employs the\npredictive covariance matrix for selection of data batches to exploit full\ncorrelation of samples. We theoretically analyse our proposed algorithm taking\ndifferent optimality criteria into consideration and provide empirical\ncomparisons highlighting the advantage of incorporating derivatives\ninformation. Our results show the effectiveness of our approach across diverse\napplications.\n", "date": "2024-08-03", "categories": []}
{"id": "2312.15357", "title": "Optimal Decision Tree and Adaptive Submodular Ranking with Noisy\n  Outcomes", "abstract": "  In pool-based active learning, the learner is given an unlabeled data set and\naims to efficiently learn the unknown hypothesis by querying the labels of the\ndata points. This can be formulated as the classical Optimal Decision Tree\n(ODT) problem: Given a set of tests, a set of hypotheses, and an outcome for\neach pair of test and hypothesis, our objective is to find a low-cost testing\nprocedure (i.e., decision tree) that identifies the true hypothesis. This\noptimization problem has been extensively studied under the assumption that\neach test generates a deterministic outcome. However, in numerous applications,\nfor example, clinical trials, the outcomes may be uncertain, which renders the\nideas from the deterministic setting invalid. In this work, we study a\nfundamental variant of the ODT problem in which some test outcomes are noisy,\neven in the more general case where the noise is persistent, i.e., repeating a\ntest gives the same noisy output. Our approximation algorithms provide\nguarantees that are nearly best possible and hold for the general case of a\nlarge number of noisy outcomes per test or per hypothesis where the performance\ndegrades continuously with this number. We numerically evaluated our algorithms\nfor identifying toxic chemicals and learning linear classifiers, and observed\nthat our algorithms have costs very close to the information-theoretic minimum.\n", "date": "2023-12-23", "categories": []}
{"id": "2301.11856", "title": "ActiveLab: Active Learning with Re-Labeling by Multiple Annotators", "abstract": "  In real-world data labeling applications, annotators often provide imperfect\nlabels. It is thus common to employ multiple annotators to label data with some\noverlap between their examples. We study active learning in such settings,\naiming to train an accurate classifier by collecting a dataset with the fewest\ntotal annotations. Here we propose ActiveLab, a practical method to decide what\nto label next that works with any classifier model and can be used in\npool-based batch active learning with one or multiple annotators. ActiveLab\nautomatically estimates when it is more informative to re-label examples vs.\nlabeling entirely new ones. This is a key aspect of producing high quality\nlabels and trained models within a limited annotation budget. In experiments on\nimage and tabular data, ActiveLab reliably trains more accurate classifiers\nwith far fewer annotations than a wide variety of popular active learning\nmethods.\n", "date": "2023-01-27", "categories": []}
{"id": "2407.18899", "title": "Learn from the Learnt: Source-Free Active Domain Adaptation via\n  Contrastive Sampling and Visual Persistence", "abstract": "  Domain Adaptation (DA) facilitates knowledge transfer from a source domain to\na related target domain. This paper investigates a practical DA paradigm,\nnamely Source data-Free Active Domain Adaptation (SFADA), where source data\nbecomes inaccessible during adaptation, and a minimum amount of annotation\nbudget is available in the target domain. Without referencing the source data,\nnew challenges emerge in identifying the most informative target samples for\nlabeling, establishing cross-domain alignment during adaptation, and ensuring\ncontinuous performance improvements through the iterative query-and-adaptation\nprocess. In response, we present learn from the learnt (LFTL), a novel paradigm\nfor SFADA to leverage the learnt knowledge from the source pretrained model and\nactively iterated models without extra overhead. We propose Contrastive Active\nSampling to learn from the hypotheses of the preceding model, thereby querying\ntarget samples that are both informative to the current model and persistently\nchallenging throughout active learning. During adaptation, we learn from\nfeatures of actively selected anchors obtained from previous intermediate\nmodels, so that the Visual Persistence-guided Adaptation can facilitate feature\ndistribution alignment and active sample exploitation. Extensive experiments on\nthree widely-used benchmarks show that our LFTL achieves state-of-the-art\nperformance, superior computational efficiency and continuous improvements as\nthe annotation budget increases. Our code is available at\nhttps://github.com/lyumengyao/lftl.\n", "date": "2024-07-26", "categories": []}
{"id": "2311.02879", "title": "Exploring Active Learning in Meta-Learning: Enhancing Context Set\n  Labeling", "abstract": "  Most meta-learning methods assume that the (very small) context set used to\nestablish a new task at test time is passively provided. In some settings,\nhowever, it is feasible to actively select which points to label; the potential\ngain from a careful choice is substantial, but the setting requires major\ndifferences from typical active learning setups. We clarify the ways in which\nactive meta-learning can be used to label a context set, depending on which\nparts of the meta-learning process use active learning. Within this framework,\nwe propose a natural algorithm based on fitting Gaussian mixtures for selecting\nwhich points to label; though simple, the algorithm also has theoretical\nmotivation. The proposed algorithm outperforms state-of-the-art active learning\nmethods when used with various meta-learning algorithms across several\nbenchmark datasets.\n", "date": "2023-11-06", "categories": []}
{"id": "2407.12212", "title": "Generalized Coverage for More Robust Low-Budget Active Learning", "abstract": "  The ProbCover method of Yehuda et al. is a well-motivated algorithm for\nactive learning in low-budget regimes, which attempts to \"cover\" the data\ndistribution with balls of a given radius at selected data points. We\ndemonstrate, however, that the performance of this algorithm is extremely\nsensitive to the choice of this radius hyper-parameter, and that tuning it is\nquite difficult, with the original heuristic frequently failing. We thus\nintroduce (and theoretically motivate) a generalized notion of \"coverage,\"\nincluding ProbCover's objective as a special case, but also allowing smoother\nnotions that are far more robust to hyper-parameter choice. We propose an\nefficient greedy method to optimize this coverage, generalizing ProbCover's\nalgorithm; due to its close connection to kernel herding, we call it\n\"MaxHerding.\" The objective can also be optimized non-greedily through a\nvariant of $k$-medoids, clarifying the relationship to other low-budget active\nlearning methods. In comprehensive experiments, MaxHerding surpasses existing\nactive learning methods across multiple low-budget image classification\nbenchmarks, and does so with less computational cost than most competitive\nmethods.\n", "date": "2024-07-16", "categories": []}
{"id": "2407.17284", "title": "A Novel Two-Step Fine-Tuning Pipeline for Cold-Start Active Learning in\n  Text Classification Tasks", "abstract": "  This is the first work to investigate the effectiveness of BERT-based\ncontextual embeddings in active learning (AL) tasks on cold-start scenarios,\nwhere traditional fine-tuning is infeasible due to the absence of labeled data.\nOur primary contribution is the proposal of a more robust fine-tuning pipeline\n- DoTCAL - that diminishes the reliance on labeled data in AL using two steps:\n(1) fully leveraging unlabeled data through domain adaptation of the embeddings\nvia masked language modeling and (2) further adjusting model weights using\nlabeled data selected by AL. Our evaluation contrasts BERT-based embeddings\nwith other prevalent text representation paradigms, including Bag of Words\n(BoW), Latent Semantic Indexing (LSI), and FastText, at two critical stages of\nthe AL process: instance selection and classification. Experiments conducted on\neight ATC benchmarks with varying AL budgets (number of labeled instances) and\nnumber of instances (about 5,000 to 300,000) demonstrate DoTCAL's superior\neffectiveness, achieving up to a 33% improvement in Macro-F1 while reducing\nlabeling efforts by half compared to the traditional one-step method. We also\nfound that in several tasks, BoW and LSI (due to information aggregation)\nproduce results superior (up to 59% ) to BERT, especially in low-budget\nscenarios and hard-to-classify tasks, which is quite surprising.\n", "date": "2024-07-24", "categories": []}
{"id": "2310.15411", "title": "Efficient Active Learning Halfspaces with Tsybakov Noise: A Non-convex\n  Optimization Approach", "abstract": "  We study the problem of computationally and label efficient PAC active\nlearning $d$-dimensional halfspaces with Tsybakov\nNoise~\\citep{tsybakov2004optimal} under structured unlabeled data\ndistributions. Inspired by~\\cite{diakonikolas2020learning}, we prove that any\napproximate first-order stationary point of a smooth nonconvex loss function\nyields a halfspace with a low excess error guarantee. In light of the above\nstructural result, we design a nonconvex optimization-based algorithm with a\nlabel complexity of $\\tilde{O}(d\n(\\frac{1}{\\epsilon})^{\\frac{8-6\\alpha}{3\\alpha-1}})$, under the assumption that\nthe Tsybakov noise parameter $\\alpha \\in (\\frac13, 1]$, which narrows down the\ngap between the label complexities of the previously known efficient passive or\nactive algorithms~\\citep{diakonikolas2020polynomial,zhang2021improved} and the\ninformation-theoretic lower bound in this setting.\n", "date": "2023-10-23", "categories": []}
{"id": "2407.14720", "title": "Downstream-Pretext Domain Knowledge Traceback for Active Learning", "abstract": "  Active learning (AL) is designed to construct a high-quality labeled dataset\nby iteratively selecting the most informative samples. Such sampling heavily\nrelies on data representation, while recently pre-training is popular for\nrobust feature learning. However, as pre-training utilizes low-level pretext\ntasks that lack annotation, directly using pre-trained representation in AL is\ninadequate for determining the sampling score. To address this problem, we\npropose a downstream-pretext domain knowledge traceback (DOKT) method that\ntraces the data interactions of downstream knowledge and pre-training guidance\nfor selecting diverse and instructive samples near the decision boundary. DOKT\nconsists of a traceback diversity indicator and a domain-based uncertainty\nestimator. The diversity indicator constructs two feature spaces based on the\npre-training pretext model and the downstream knowledge from annotation, by\nwhich it locates the neighbors of unlabeled data from the downstream space in\nthe pretext space to explore the interaction of samples. With this mechanism,\nDOKT unifies the data relations of low-level and high-level representations to\nestimate traceback diversity. Next, in the uncertainty estimator, domain mixing\nis designed to enforce perceptual perturbing to unlabeled samples with similar\nvisual patches in the pretext space. Then the divergence of perturbed samples\nis measured to estimate the domain uncertainty. As a result, DOKT selects the\nmost diverse and important samples based on these two modules. The experiments\nconducted on ten datasets show that our model outperforms other\nstate-of-the-art methods and generalizes well to various application scenarios\nsuch as semantic segmentation and image captioning.\n", "date": "2024-07-20", "categories": []}
{"id": "2407.13364", "title": "Geometric Active Exploration in Markov Decision Processes: the Benefit\n  of Abstraction", "abstract": "  How can a scientist use a Reinforcement Learning (RL) algorithm to design\nexperiments over a dynamical system's state space? In the case of finite and\nMarkovian systems, an area called Active Exploration (AE) relaxes the\noptimization problem of experiments design into Convex RL, a generalization of\nRL admitting a wider notion of reward. Unfortunately, this framework is\ncurrently not scalable and the potential of AE is hindered by the vastness of\nexperiment spaces typical of scientific discovery applications. However, these\nspaces are often endowed with natural geometries, e.g., permutation invariance\nin molecular design, that an agent could leverage to improve the statistical\nand computational efficiency of AE. To achieve this, we bridge AE and MDP\nhomomorphisms, which offer a way to exploit known geometric structures via\nabstraction. Towards this goal, we make two fundamental contributions: we\nextend MDP homomorphisms formalism to Convex RL, and we present, to the best of\nour knowledge, the first analysis that formally captures the benefit of\nabstraction via homomorphisms on sample efficiency. Ultimately, we propose the\nGeometric Active Exploration (GAE) algorithm, which we analyse theoretically\nand experimentally in environments motivated by problems in scientific\ndiscovery.\n", "date": "2024-07-18", "categories": []}
{"id": "2407.13429", "title": "Towards Dynamic Feature Acquisition on Medical Time Series by Maximizing\n  Conditional Mutual Information", "abstract": "  Knowing which features of a multivariate time series to measure and when is a\nkey task in medicine, wearables, and robotics. Better acquisition policies can\nreduce costs while maintaining or even improving the performance of downstream\npredictors. Inspired by the maximization of conditional mutual information, we\npropose an approach to train acquirers end-to-end using only the downstream\nloss. We show that our method outperforms random acquisition policy, matches a\nmodel with an unrestrained budget, but does not yet overtake a static\nacquisition strategy. We highlight the assumptions and outline avenues for\nfuture work.\n", "date": "2024-07-18", "categories": []}
{"id": "2407.11053", "title": "Sampling and active learning methods for network reliability estimation\n  using K-terminal spanning tree", "abstract": "  Network reliability analysis remains a challenge due to the increasing size\nand complexity of networks. This paper presents a novel sampling method and an\nactive learning method for efficient and accurate network reliability\nestimation under node failure and edge failure scenarios. The proposed sampling\nmethod adopts Monte Carlo technique to sample component lifetimes and the\nK-terminal spanning tree algorithm to accelerate structure function\ncomputation. Unlike existing methods that compute only one structure function\nvalue per sample, our method generates multiple component state vectors and\ncorresponding structure function values from each sample. Network reliability\nis estimated based on survival signatures derived from these values. A\ntransformation technique extends this method to handle both node failure and\nedge failure. To enhance efficiency of proposed sampling method and achieve\nadaptability to network topology changes, we introduce an active learning\nmethod utilizing a random forest (RF) classifier. This classifier directly\npredicts structure function values, integrates network behaviors across diverse\ntopologies, and undergoes iterative refinement to enhance predictive accuracy.\nImportantly, the trained RF classifier can directly predict reliability for\nvariant networks, a capability beyond the sampling method alone. Through\ninvestigating several network examples and two practical applications, the\neffectiveness of both proposed methods is demonstrated.\n", "date": "2024-07-09", "categories": []}
{"id": "2405.00334", "title": "A Survey on Deep Active Learning: Recent Advances and New Frontiers", "abstract": "  Active learning seeks to achieve strong performance with fewer training\nsamples. It does this by iteratively asking an oracle to label new selected\nsamples in a human-in-the-loop manner. This technique has gained increasing\npopularity due to its broad applicability, yet its survey papers, especially\nfor deep learning-based active learning (DAL), remain scarce. Therefore, we\nconduct an advanced and comprehensive survey on DAL. We first introduce\nreviewed paper collection and filtering. Second, we formally define the DAL\ntask and summarize the most influential baselines and widely used datasets.\nThird, we systematically provide a taxonomy of DAL methods from five\nperspectives, including annotation types, query strategies, deep model\narchitectures, learning paradigms, and training processes, and objectively\nanalyze their strengths and weaknesses. Then, we comprehensively summarize main\napplications of DAL in Natural Language Processing (NLP), Computer Vision (CV),\nand Data Mining (DM), etc. Finally, we discuss challenges and perspectives\nafter a detailed analysis of current studies. This work aims to serve as a\nuseful and quick guide for researchers in overcoming difficulties in DAL. We\nhope that this survey will spur further progress in this burgeoning field.\n", "date": "2024-05-01", "categories": []}
{"id": "2407.10196", "title": "A3S: A General Active Clustering Method with Pairwise Constraints", "abstract": "  Active clustering aims to boost the clustering performance by integrating\nhuman-annotated pairwise constraints through strategic querying. Conventional\napproaches with semi-supervised clustering schemes encounter high query costs\nwhen applied to large datasets with numerous classes. To address these\nlimitations, we propose a novel Adaptive Active Aggregation and Splitting (A3S)\nframework, falling within the cluster-adjustment scheme in active clustering.\nA3S features strategic active clustering adjustment on the initial cluster\nresult, which is obtained by an adaptive clustering algorithm. In particular,\nour cluster adjustment is inspired by the quantitative analysis of Normalized\nmutual information gain under the information theory framework and can provably\nimprove the clustering quality. The proposed A3S framework significantly\nelevates the performance and scalability of active clustering. In extensive\nexperiments across diverse real-world datasets, A3S achieves desired results\nwith significantly fewer human queries compared with existing methods.\n", "date": "2024-07-14", "categories": []}
{"id": "2402.13468", "title": "STENCIL: Submodular Mutual Information Based Weak Supervision for\n  Cold-Start Active Learning", "abstract": "  As supervised fine-tuning of pre-trained models within NLP applications\nincreases in popularity, larger corpora of annotated data are required,\nespecially with increasing parameter counts in large language models. Active\nlearning, which attempts to mine and annotate unlabeled instances to improve\nmodel performance maximally fast, is a common choice for reducing the\nannotation cost; however, most methods typically ignore class imbalance and\neither assume access to initial annotated data or require multiple rounds of\nactive learning selection before improving rare classes. We present STENCIL,\nwhich utilizes a set of text exemplars and the recently proposed submodular\nmutual information to select a set of weakly labeled rare-class instances that\nare then strongly labeled by an annotator. We show that STENCIL improves\noverall accuracy by $10\\%-18\\%$ and rare-class F-1 score by $17\\%-40\\%$ on\nmultiple text classification datasets over common active learning methods\nwithin the class-imbalanced cold-start setting.\n", "date": "2024-02-21", "categories": []}
{"id": "2405.09312", "title": "Agnostic Active Learning of Single Index Models with Linear Sample\n  Complexity", "abstract": "  We study active learning methods for single index models of the form\n$F({\\mathbf x}) = f(\\langle {\\mathbf w}, {\\mathbf x}\\rangle)$, where\n$f:\\mathbb{R} \\to \\mathbb{R}$ and ${\\mathbf x,\\mathbf w} \\in \\mathbb{R}^d$. In\naddition to their theoretical interest as simple examples of non-linear neural\nnetworks, single index models have received significant recent attention due to\napplications in scientific machine learning like surrogate modeling for partial\ndifferential equations (PDEs). Such applications require sample-efficient\nactive learning methods that are robust to adversarial noise. I.e., that work\neven in the challenging agnostic learning setting.\n  We provide two main results on agnostic active learning of single index\nmodels. First, when $f$ is known and Lipschitz, we show that $\\tilde{O}(d)$\nsamples collected via {statistical leverage score sampling} are sufficient to\nlearn a near-optimal single index model. Leverage score sampling is simple to\nimplement, efficient, and already widely used for actively learning linear\nmodels. Our result requires no assumptions on the data distribution, is optimal\nup to log factors, and improves quadratically on a recent ${O}(d^{2})$ bound of\n\\cite{gajjar2023active}. Second, we show that $\\tilde{O}(d)$ samples suffice\neven in the more difficult setting when $f$ is \\emph{unknown}. Our results\nleverage tools from high dimensional probability, including Dudley's inequality\nand dual Sudakov minoration, as well as a novel, distribution-aware\ndiscretization of the class of Lipschitz functions.\n", "date": "2024-05-15", "categories": []}
{"id": "2208.07497", "title": "Bucketized Active Sampling for Learning ACOPF", "abstract": "  This paper considers optimization proxies for Optimal Power Flow (OPF), i.e.,\nmachine-learning models that approximate the input/output relationship of OPF.\nRecent work has focused on showing that such proxies can be of high fidelity.\nHowever, their training requires significant data, each instance necessitating\nthe (offline) solving of an OPF. To meet the requirements of market-clearing\napplications, this paper proposes Bucketized Active Sampling (BAS), a novel\nactive learning framework that aims at training the best possible OPF proxy\nwithin a time limit. BAS partitions the input domain into buckets and uses an\nacquisition function to determine where to sample next. By applying the same\npartitioning to the validation set, BAS leverages labeled validation samples in\nthe selection of unlabeled samples. BAS also relies on an adaptive learning\nrate that increases and decreases over time. Experimental results demonstrate\nthe benefits of BAS.\n", "date": "2022-08-16", "categories": []}
{"id": "2303.11530", "title": "Active Coarse-to-Fine Segmentation of Moveable Parts from Real Images", "abstract": "  We introduce the first active learning (AL) model for high-accuracy instance\nsegmentation of moveable parts from RGB images of real indoor scenes.\nSpecifically, our goal is to obtain fully validated segmentation results by\nhumans while minimizing manual effort. To this end, we employ a transformer\nthat utilizes a masked-attention mechanism to supervise the active\nsegmentation. To enhance the network tailored to moveable parts, we introduce a\ncoarse-to-fine AL approach which first uses an object-aware masked attention\nand then a pose-aware one, leveraging the hierarchical nature of the problem\nand a correlation between moveable parts and object poses and interaction\ndirections. When applying our AL model to 2,000 real images, we obtain fully\nvalidated moveable part segmentations with semantic labels, by only needing to\nmanually annotate 11.45% of the images. This translates to significant (60%)\ntime saving over manual effort required by the best non-AL model to attain the\nsame segmentation accuracy. At last, we contribute a dataset of 2,550 real\nimages with annotated moveable parts, demonstrating its superior quality and\ndiversity over the best alternatives.\n", "date": "2023-03-21", "categories": []}
{"id": "2307.14380", "title": "Robust Assignment of Labels for Active Learning with Sparse and Noisy\n  Annotations", "abstract": "  Supervised classification algorithms are used to solve a growing number of\nreal-life problems around the globe. Their performance is strictly connected\nwith the quality of labels used in training. Unfortunately, acquiring\ngood-quality annotations for many tasks is infeasible or too expensive to be\ndone in practice. To tackle this challenge, active learning algorithms are\ncommonly employed to select only the most relevant data for labeling. However,\nthis is possible only when the quality and quantity of labels acquired from\nexperts are sufficient. Unfortunately, in many applications, a trade-off\nbetween annotating individual samples by multiple annotators to increase label\nquality vs. annotating new samples to increase the total number of labeled\ninstances is necessary. In this paper, we address the issue of faulty data\nannotations in the context of active learning. In particular, we propose two\nnovel annotation unification algorithms that utilize unlabeled parts of the\nsample space. The proposed methods require little to no intersection between\nsamples annotated by different experts. Our experiments on four public datasets\nindicate the robustness and superiority of the proposed methods in both, the\nestimation of the annotator's reliability, and the assignment of actual labels,\nagainst the state-of-the-art algorithms and the simple majority voting.\n", "date": "2023-07-25", "categories": []}
{"id": "2407.02335", "title": "CALICO: Confident Active Learning with Integrated Calibration", "abstract": "  The growing use of deep learning in safety-critical applications, such as\nmedical imaging, has raised concerns about limited labeled data, where this\ndemand is amplified as model complexity increases, posing hurdles for domain\nexperts to annotate data. In response to this, active learning (AL) is used to\nefficiently train models with limited annotation costs. In the context of deep\nneural networks (DNNs), AL often uses confidence or probability outputs as a\nscore for selecting the most informative samples. However, modern DNNs exhibit\nunreliable confidence outputs, making calibration essential. We propose an AL\nframework that self-calibrates the confidence used for sample selection during\nthe training process, referred to as Confident Active Learning with Integrated\nCalibratiOn (CALICO). CALICO incorporates the joint training of a classifier\nand an energy-based model, instead of the standard softmax-based classifier.\nThis approach allows for simultaneous estimation of the input data distribution\nand the class probabilities during training, improving calibration without\nneeding an additional labeled dataset. Experimental results showcase improved\nclassification performance compared to a softmax-based classifier with fewer\nlabeled samples. Furthermore, the calibration stability of the model is\nobserved to depend on the prior class distribution of the data.\n", "date": "2024-07-02", "categories": []}
{"id": "2308.13201", "title": "Deep Active Audio Feature Learning in Resource-Constrained Environments", "abstract": "  The scarcity of labelled data makes training Deep Neural Network (DNN) models\nin bioacoustic applications challenging. In typical bioacoustics applications,\nmanually labelling the required amount of data can be prohibitively expensive.\nTo effectively identify both new and current classes, DNN models must continue\nto learn new features from a modest amount of fresh data. Active Learning (AL)\nis an approach that can help with this learning while requiring little\nlabelling effort. Nevertheless, the use of fixed feature extraction approaches\nlimits feature quality, resulting in underutilization of the benefits of AL. We\ndescribe an AL framework that addresses this issue by incorporating feature\nextraction into the AL loop and refining the feature extractor after each round\nof manual annotation. In addition, we use raw audio processing rather than\nspectrograms, which is a novel approach. Experiments reveal that the proposed\nAL framework requires 14.3%, 66.7%, and 47.4% less labelling effort on\nbenchmark audio datasets ESC-50, UrbanSound8k, and InsectWingBeat,\nrespectively, for a large DNN model and similar savings on a\nmicrocontroller-based counterpart. Furthermore, we showcase the practical\nrelevance of our study by incorporating data from conservation biology\nprojects. All codes are publicly available on GitHub.\n", "date": "2023-08-25", "categories": []}
{"id": "2301.12616", "title": "Active Sequential Two-Sample Testing", "abstract": "  A two-sample hypothesis test is a statistical procedure used to determine\nwhether the distributions generating two samples are identical. We consider the\ntwo-sample testing problem in a new scenario where the sample measurements (or\nsample features) are inexpensive to access, but their group memberships (or\nlabels) are costly. To address the problem, we devise the first \\emph{active\nsequential two-sample testing framework} that not only sequentially but also\n\\emph{actively queries}. Our test statistic is a likelihood ratio where one\nlikelihood is found by maximization over all class priors, and the other is\nprovided by a probabilistic classification model. The classification model is\nadaptively updated and used to predict where the (unlabelled) features have a\nhigh dependency on labels; labeling the ``high-dependency'' features leads to\nthe increased power of the proposed testing framework. In theory, we provide\nthe proof that our framework produces an \\emph{anytime-valid} $p$-value. In\naddition, we characterize the proposed framework's gain in testing power by\nanalyzing the mutual information between the feature and label variables in\nasymptotic and finite-sample scenarios. In practice, we introduce an\ninstantiation of our framework and evaluate it using several experiments; the\nexperiments on the synthetic, MNIST, and application-specific datasets\ndemonstrate that the testing power of the instantiated active sequential test\nsignificantly increases while the Type I error is under control.\n", "date": "2023-01-30", "categories": []}
{"id": "2406.19714", "title": "State Matching and Multiple References in Adaptive Active Automata\n  Learning", "abstract": "  Active automata learning (AAL) is a method to infer state machines by\ninteracting with black-box systems. Adaptive AAL aims to reduce the sample\ncomplexity of AAL by incorporating domain specific knowledge in the form of\n(similar) reference models. Such reference models appear naturally when\nlearning multiple versions or variants of a software system. In this paper, we\npresent state matching, which allows flexible use of the structure of these\nreference models by the learner. State matching is the main ingredient of\nadaptive L#, a novel framework for adaptive learning, built on top of L#. Our\nempirical evaluation shows that adaptive L# improves the state of the art by up\nto two orders of magnitude.\n", "date": "2024-06-28", "categories": []}
{"id": "2312.12009", "title": "Active Preference Inference using Language Models and Probabilistic\n  Reasoning", "abstract": "  Actively inferring user preferences, for example by asking good questions, is\nimportant for any human-facing decision-making system. Active inference allows\nsuch systems to adapt and personalize themselves to nuanced individual\npreferences. To enable this ability for instruction-tuned large language models\n(LLMs), one may prompt them to ask users questions to infer their preferences,\ntransforming the language models into more robust, interactive systems.\nHowever, out of the box, these models are not efficient at extracting\npreferences: the questions they generate are not informative, requiring a high\nnumber of user interactions and impeding the usability of the downstream\nsystem. In this work, we introduce an inference-time algorithm that helps LLMs\nquickly infer preferences by using more informative questions. Our algorithm\nuses a probabilistic model whose conditional distributions are defined by\nprompting an LLM, and returns questions that optimize expected entropy and\nexpected model change. Results in a simplified interactive web shopping setting\nwith real product items show that an LLM equipped with our entropy reduction\nalgorithm outperforms baselines with the same underlying LLM on task\nperformance while using fewer user interactions.\n", "date": "2023-12-19", "categories": []}
{"id": "2401.14555", "title": "Revisiting Active Learning in the Era of Vision Foundation Models", "abstract": "  Foundation vision or vision-language models are trained on large unlabeled or\nnoisy data and learn robust representations that can achieve impressive zero-\nor few-shot performance on diverse tasks. Given these properties, they are a\nnatural fit for active learning (AL), which aims to maximize labeling\nefficiency. However, the full potential of foundation models has not been\nexplored in the context of AL, specifically in the low-budget regime. In this\nwork, we evaluate how foundation models influence three critical components of\neffective AL, namely, 1) initial labeled pool selection, 2) ensuring diverse\nsampling, and 3) the trade-off between representative and uncertainty sampling.\nWe systematically study how the robust representations of foundation models\n(DINOv2, OpenCLIP) challenge existing findings in active learning. Our\nobservations inform the principled construction of a new simple and elegant AL\nstrategy that balances uncertainty estimated via dropout with sample diversity.\nWe extensively test our strategy on many challenging image classification\nbenchmarks, including natural images as well as out-of-domain biomedical images\nthat are relatively understudied in the AL literature. We also provide a highly\nperformant and efficient implementation of modern AL strategies (including our\nmethod) at https://github.com/sanketx/AL-foundation-models.\n", "date": "2024-01-25", "categories": []}
{"id": "2405.13285", "title": "Enhancing Active Learning for Sentinel 2 Imagery through Contrastive\n  Learning and Uncertainty Estimation", "abstract": "  In this paper, we introduce a novel method designed to enhance label\nefficiency in satellite imagery analysis by integrating semi-supervised\nlearning (SSL) with active learning strategies. Our approach utilizes\ncontrastive learning together with uncertainty estimations via Monte Carlo\nDropout (MC Dropout), with a particular focus on Sentinel-2 imagery analyzed\nusing the Eurosat dataset. We explore the effectiveness of our method in\nscenarios featuring both balanced and unbalanced class distributions. Our\nresults show that the proposed method performs better than several other\npopular methods in this field, enabling significant savings in labeling effort\nwhile maintaining high classification accuracy. These findings highlight the\npotential of our approach to facilitate scalable and cost-effective satellite\nimage analysis, particularly advantageous for extensive environmental\nmonitoring and land use classification tasks.\n", "date": "2024-05-22", "categories": []}
{"id": "2406.17322", "title": "ALPBench: A Benchmark for Active Learning Pipelines on Tabular Data", "abstract": "  In settings where only a budgeted amount of labeled data can be afforded,\nactive learning seeks to devise query strategies for selecting the most\ninformative data points to be labeled, aiming to enhance learning algorithms'\nefficiency and performance. Numerous such query strategies have been proposed\nand compared in the active learning literature. However, the community still\nlacks standardized benchmarks for comparing the performance of different query\nstrategies. This particularly holds for the combination of query strategies\nwith different learning algorithms into active learning pipelines and examining\nthe impact of the learning algorithm choice. To close this gap, we propose\nALPBench, which facilitates the specification, execution, and performance\nmonitoring of active learning pipelines. It has built-in measures to ensure\nevaluations are done reproducibly, saving exact dataset splits and\nhyperparameter settings of used algorithms. In total, ALPBench consists of 86\nreal-world tabular classification datasets and 5 active learning settings,\nyielding 430 active learning problems. To demonstrate its usefulness and broad\ncompatibility with various learning algorithms and query strategies, we conduct\nan exemplary study evaluating 9 query strategies paired with 8 learning\nalgorithms in 2 different settings. We provide ALPBench here:\nhttps://github.com/ValentinMargraf/ActiveLearningPipelines.\n", "date": "2024-06-25", "categories": []}
{"id": "2406.12560", "title": "Towards Bayesian Data Selection", "abstract": "  A wide range of machine learning algorithms iteratively add data to the\ntraining sample. Examples include semi-supervised learning, active learning,\nmulti-armed bandits, and Bayesian optimization. We embed this kind of data\naddition into decision theory by framing data selection as a decision problem.\nThis paves the way for finding Bayes-optimal selections of data. For the\nillustrative case of self-training in semi-supervised learning, we derive the\nrespective Bayes criterion. We further show that deploying this criterion\nmitigates the issue of confirmation bias by empirically assessing our method\nfor generalized linear models, semi-parametric generalized additive models, and\nBayesian neural networks on simulated and real-world data.\n", "date": "2024-06-18", "categories": []}
{"id": "2402.15441", "title": "Active Few-Shot Fine-Tuning", "abstract": "  We study the question: How can we select the right data for fine-tuning to a\nspecific task? We call this data selection problem active fine-tuning and show\nthat it is an instance of transductive active learning, a novel generalization\nof classical active learning. We propose ITL, short for information-based\ntransductive learning, an approach which samples adaptively to maximize\ninformation gained about the specified task. We are the first to show, under\ngeneral regularity assumptions, that such decision rules converge uniformly to\nthe smallest possible uncertainty obtainable from the accessible data. We apply\nITL to the few-shot fine-tuning of large neural networks and show that\nfine-tuning with ITL learns the task with significantly fewer examples than the\nstate-of-the-art.\n", "date": "2024-02-13", "categories": []}
{"id": "2406.14784", "title": "Active Learning for Fair and Stable Online Allocations", "abstract": "  We explore an active learning approach for dynamic fair resource allocation\nproblems. Unlike previous work that assumes full feedback from all agents on\ntheir allocations, we consider feedback from a select subset of agents at each\nepoch of the online resource allocation process. Despite this restriction, our\nproposed algorithms provide regret bounds that are sub-linear in number of\ntime-periods for various measures that include fairness metrics commonly used\nin resource allocation problems and stability considerations in matching\nmechanisms. The key insight of our algorithms lies in adaptively identifying\nthe most informative feedback using dueling upper and lower confidence bounds.\nWith this strategy, we show that efficient decision-making does not require\nextensive feedback and produces efficient outcomes for a variety of problem\nclasses.\n", "date": "2024-06-20", "categories": []}
{"id": "2405.02612", "title": "Learning Linear Utility Functions From Pairwise Comparison Queries", "abstract": "  We study learnability of linear utility functions from pairwise comparison\nqueries. In particular, we consider two learning objectives. The first\nobjective is to predict out-of-sample responses to pairwise comparisons,\nwhereas the second is to approximately recover the true parameters of the\nutility function. We show that in the passive learning setting, linear\nutilities are efficiently learnable with respect to the first objective, both\nwhen query responses are uncorrupted by noise, and under Tsybakov noise when\nthe distributions are sufficiently \"nice\". In contrast, we show that utility\nparameters are not learnable for a large set of data distributions without\nstrong modeling assumptions, even when query responses are noise-free. Next, we\nproceed to analyze the learning problem in an active learning setting. In this\ncase, we show that even the second objective is efficiently learnable, and\npresent algorithms for both the noise-free and noisy query response settings.\nOur results thus exhibit a qualitative learnability gap between passive and\nactive learning from pairwise preference queries, demonstrating the value of\nthe ability to select pairwise queries for utility learning.\n", "date": "2024-05-04", "categories": []}
{"id": "2406.12114", "title": "Enhancing Text Classification through LLM-Driven Active Learning and\n  Human Annotation", "abstract": "  In the context of text classification, the financial burden of annotation\nexercises for creating training data is a critical issue. Active learning\ntechniques, particularly those rooted in uncertainty sampling, offer a\ncost-effective solution by pinpointing the most instructive samples for manual\nannotation. Similarly, Large Language Models (LLMs) such as GPT-3.5 provide an\nalternative for automated annotation but come with concerns regarding their\nreliability. This study introduces a novel methodology that integrates human\nannotators and LLMs within an Active Learning framework. We conducted\nevaluations on three public datasets. IMDB for sentiment analysis, a Fake News\ndataset for authenticity discernment, and a Movie Genres dataset for\nmulti-label classification.The proposed framework integrates human annotation\nwith the output of LLMs, depending on the model uncertainty levels. This\nstrategy achieves an optimal balance between cost efficiency and classification\nperformance. The empirical results show a substantial decrease in the costs\nassociated with data annotation while either maintaining or improving model\naccuracy.\n", "date": "2024-06-17", "categories": []}
{"id": "2406.11310", "title": "Federated Active Learning Framework for Efficient Annotation Strategy in\n  Skin-lesion Classification", "abstract": "  Federated Learning (FL) enables multiple institutes to train models\ncollaboratively without sharing private data. Current FL research focuses on\ncommunication efficiency, privacy protection, and personalization and assumes\nthat the data of FL have already been ideally collected. In medical scenarios,\nhowever, data annotation demands both expertise and intensive labor, which is a\ncritical problem in FL. Active learning (AL), has shown promising performance\nin reducing the number of data annotations in medical image analysis. We\npropose a federated AL (FedAL) framework in which AL is executed periodically\nand interactively under FL. We exploit a local model in each hospital and a\nglobal model acquired from FL to construct an ensemble. We use\nensemble-entropy-based AL as an efficient data-annotation strategy in FL.\nTherefore, our FedAL framework can decrease the amount of annotated data and\npreserve patient privacy while maintaining the performance of FL. To our\nknowledge, this is the first FedAL framework applied to medical images. We\nvalidated our framework on real-world dermoscopic datasets. Using only 50% of\nsamples, our framework was able to achieve state-of-the-art performance on a\nskin-lesion classification task. Our framework performed better than several\nstate-of-the-art AL methods under FL and achieved comparable performance to\nfull-data FL.\n", "date": "2024-06-17", "categories": []}
{"id": "2406.11485", "title": "Active clustering with bandit feedback", "abstract": "  We investigate the Active Clustering Problem (ACP). A learner interacts with\nan $N$-armed stochastic bandit with $d$-dimensional subGaussian feedback. There\nexists a hidden partition of the arms into $K$ groups, such that arms within\nthe same group, share the same mean vector. The learner's task is to uncover\nthis hidden partition with the smallest budget - i.e., the least number of\nobservation - and with a probability of error smaller than a prescribed\nconstant $\\delta$. In this paper, (i) we derive a non-asymptotic lower bound\nfor the budget, and (ii) we introduce the computationally efficient ACB\nalgorithm, whose budget matches the lower bound in most regimes. We improve on\nthe performance of a uniform sampling strategy. Importantly, contrary to the\nbatch setting, we establish that there is no computation-information gap in the\nactive setting.\n", "date": "2024-06-17", "categories": []}
{"id": "2406.09296", "title": "Parameter-Efficient Active Learning for Foundational models", "abstract": "  Foundational vision transformer models have shown impressive few shot\nperformance on many vision tasks. This research presents a novel investigation\ninto the application of parameter efficient fine-tuning methods within an\nactive learning (AL) framework, to advance the sampling selection process in\nextremely budget constrained classification tasks. The focus on image datasets,\nknown for their out-of-distribution characteristics, adds a layer of complexity\nand relevance to our study. Through a detailed evaluation, we illustrate the\nimproved AL performance on these challenging datasets, highlighting the\nstrategic advantage of merging parameter efficient fine tuning methods with\nfoundation models. This contributes to the broader discourse on optimizing AL\nstrategies, presenting a promising avenue for future exploration in leveraging\nfoundation models for efficient and effective data annotation in specialized\ndomains.\n", "date": "2024-06-13", "categories": []}
{"id": "2309.12443", "title": "Active Learning for Multilingual Fingerspelling Corpora", "abstract": "  We apply active learning to help with data scarcity problems in sign\nlanguages. In particular, we perform a novel analysis of the effect of\npre-training. Since many sign languages are linguistic descendants of French\nsign language, they share hand configurations, which pre-training can hopefully\nexploit. We test this hypothesis on American, Chinese, German, and Irish\nfingerspelling corpora. We do observe a benefit from pre-training, but this may\nbe due to visual rather than linguistic similarities\n", "date": "2023-09-21", "categories": []}
{"id": "2405.07937", "title": "Active Learning with Simple Questions", "abstract": "  We consider an active learning setting where a learner is presented with a\npool S of n unlabeled examples belonging to a domain X and asks queries to find\nthe underlying labeling that agrees with a target concept h^* \\in H.\n  In contrast to traditional active learning that queries a single example for\nits label, we study more general region queries that allow the learner to pick\na subset of the domain T \\subset X and a target label y and ask a labeler\nwhether h^*(x) = y for every example in the set T \\cap S.\n  Such more powerful queries allow us to bypass the limitations of traditional\nactive learning and use significantly fewer rounds of interactions to learn but\ncan potentially lead to a significantly more complex query language. Our main\ncontribution is quantifying the trade-off between the number of queries and the\ncomplexity of the query language used by the learner.\n  We measure the complexity of the region queries via the VC dimension of the\nfamily of regions. We show that given any hypothesis class H with VC dimension\nd, one can design a region query family Q with VC dimension O(d) such that for\nevery set of n examples S \\subset X and every h^* \\in H, a learner can submit\nO(d log n) queries from Q to a labeler and perfectly label S. We show a\nmatching lower bound by designing a hypothesis class H with VC dimension d and\na dataset S \\subset X of size n such that any learning algorithm using any\nquery class with VC dimension less than O(d) must make poly(n) queries to label\nS perfectly.\n  Finally, we focus on well-studied hypothesis classes including unions of\nintervals, high-dimensional boxes, and d-dimensional halfspaces, and obtain\nstronger results. In particular, we design learning algorithms that (i) are\ncomputationally efficient and (ii) work even when the queries are not answered\nbased on the learner's pool of examples S but on some unknown superset L of S\n", "date": "2024-05-13", "categories": []}
{"id": "2406.06061", "title": "Greedy SLIM: A SLIM-Based Approach For Preference Elicitation", "abstract": "  Preference elicitation is an active learning approach to tackle the\ncold-start problem of recommender systems. Roughly speaking, new users are\nasked to rate some carefully selected items in order to compute appropriate\nrecommendations for them. To the best of our knowledge, we are the first to\npropose a method for preference elicitation that is based on SLIM , a\nstate-of-the-art technique for top-N recommendation. Our approach mainly\nconsists of a new training technique for SLIM, which we call Greedy SLIM. This\ntechnique iteratively selects items for the training in order to minimize the\nSLIM loss greedily. We conduct offline experiments as well as a user study to\nassess the performance of this new method. The results are remarkable,\nespecially with respect to the user study. We conclude that Greedy SLIM seems\nto be more suitable for preference elicitation than widely used methods based\non latent factor models.\n", "date": "2024-06-10", "categories": []}
{"id": "2406.03630", "title": "Active ML for 6G: Towards Efficient Data Generation, Acquisition, and\n  Annotation", "abstract": "  This paper explores the integration of active machine learning (ML) for 6G\nnetworks, an area that remains under-explored yet holds potential. Unlike\npassive ML systems, active ML can be made to interact with the network\nenvironment. It actively selects informative and representative data points for\ntraining, thereby reducing the volume of data needed while accelerating the\nlearning process. While active learning research mainly focuses on data\nannotation, we call for a network-centric active learning framework that\nconsiders both annotation (i.e., what is the label) and data acquisition (i.e.,\nwhich and how many samples to collect). Moreover, we explore the synergy\nbetween generative artificial intelligence (AI) and active learning to overcome\nexisting limitations in both active learning and generative AI. This paper also\nfeatures a case study on a mmWave throughput prediction problem to demonstrate\nthe practical benefits and improved performance of active learning for 6G\nnetworks. Furthermore, we discuss how the implications of active learning\nextend to numerous 6G network use cases. We highlight the potential of active\nlearning based 6G networks to enhance computational efficiency, data annotation\nand acquisition efficiency, adaptability, and overall network intelligence. We\nconclude with a discussion on challenges and future research directions for\nactive learning in 6G networks, including development of novel query\nstrategies, distributed learning integration, and inclusion of human- and\nmachine-in-the-loop learning.\n", "date": "2024-06-05", "categories": []}
{"id": "2406.03944", "title": "Provably Neural Active Learning Succeeds via Prioritizing Perplexing\n  Samples", "abstract": "  Neural Network-based active learning (NAL) is a cost-effective data selection\ntechnique that utilizes neural networks to select and train on a small subset\nof samples. While existing work successfully develops various effective or\ntheory-justified NAL algorithms, the understanding of the two commonly used\nquery criteria of NAL: uncertainty-based and diversity-based, remains in its\ninfancy. In this work, we try to move one step forward by offering a unified\nexplanation for the success of both query criteria-based NAL from a feature\nlearning view. Specifically, we consider a feature-noise data model comprising\neasy-to-learn or hard-to-learn features disrupted by noise, and conduct\nanalysis over 2-layer NN-based NALs in the pool-based scenario. We provably\nshow that both uncertainty-based and diversity-based NAL are inherently\namenable to one and the same principle, i.e., striving to prioritize samples\nthat contain yet-to-be-learned features. We further prove that this shared\nprinciple is the key to their success-achieve small test error within a small\nlabeled set. Contrastingly, the strategy-free passive learning exhibits a large\ntest error due to the inadequate learning of yet-to-be-learned features,\nnecessitating resort to a significantly larger label complexity for a\nsufficient test error reduction. Experimental results validate our findings.\n", "date": "2024-06-06", "categories": []}
{"id": "2306.11180", "title": "Hyperbolic Active Learning for Semantic Segmentation under Domain Shift", "abstract": "  We introduce a hyperbolic neural network approach to pixel-level active\nlearning for semantic segmentation. Analysis of the data statistics leads to a\nnovel interpretation of the hyperbolic radius as an indicator of data scarcity.\nIn HALO (Hyperbolic Active Learning Optimization), for the first time, we\npropose the use of epistemic uncertainty as a data acquisition strategy,\nfollowing the intuition of selecting data points that are the least known. The\nhyperbolic radius, complemented by the widely-adopted prediction entropy,\neffectively approximates epistemic uncertainty. We perform extensive\nexperimental analysis based on two established synthetic-to-real benchmarks,\ni.e. GTAV $\\rightarrow$ Cityscapes and SYNTHIA $\\rightarrow$ Cityscapes.\nAdditionally, we test HALO on Cityscape $\\rightarrow$ ACDC for domain\nadaptation under adverse weather conditions, and we benchmark both\nconvolutional and attention-based backbones. HALO sets a new state-of-the-art\nin active learning for semantic segmentation under domain shift and it is the\nfirst active learning approach that surpasses the performance of supervised\ndomain adaptation while using only a small portion of labels (i.e., 1%).\n", "date": "2023-06-19", "categories": []}
{"id": "2403.10820", "title": "Active Label Correction for Semantic Segmentation with Foundation Models", "abstract": "  Training and validating models for semantic segmentation require datasets\nwith pixel-wise annotations, which are notoriously labor-intensive. Although\nuseful priors such as foundation models or crowdsourced datasets are available,\nthey are error-prone. We hence propose an effective framework of active label\ncorrection (ALC) based on a design of correction query to rectify pseudo labels\nof pixels, which in turn is more annotator-friendly than the standard one\ninquiring to classify a pixel directly according to our theoretical analysis\nand user study. Specifically, leveraging foundation models providing useful\nzero-shot predictions on pseudo labels and superpixels, our method comprises\ntwo key techniques: (i) an annotator-friendly design of correction query with\nthe pseudo labels, and (ii) an acquisition function looking ahead label\nexpansions based on the superpixels. Experimental results on PASCAL,\nCityscapes, and Kvasir-SEG datasets demonstrate the effectiveness of our ALC\nframework, outperforming prior methods for active semantic segmentation and\nlabel correction. Notably, utilizing our method, we obtained a revised dataset\nof PASCAL by rectifying errors in 2.6 million pixels in PASCAL dataset.\n", "date": "2024-03-16", "categories": []}
{"id": "2406.02435", "title": "Generative Active Learning for Long-tailed Instance Segmentation", "abstract": "  Recently, large-scale language-image generative models have gained widespread\nattention and many works have utilized generated data from these models to\nfurther enhance the performance of perception tasks. However, not all generated\ndata can positively impact downstream models, and these methods do not\nthoroughly explore how to better select and utilize generated data. On the\nother hand, there is still a lack of research oriented towards active learning\non generated data. In this paper, we explore how to perform active learning\nspecifically for generated data in the long-tailed instance segmentation task.\nSubsequently, we propose BSGAL, a new algorithm that online estimates the\ncontribution of the generated data based on gradient cache. BSGAL can handle\nunlimited generated data and complex downstream segmentation tasks effectively.\nExperiments show that BSGAL outperforms the baseline approach and effectually\nimproves the performance of long-tailed segmentation. Our code can be found at\nhttps://github.com/aim-uofa/DiverGen.\n", "date": "2024-06-04", "categories": []}
{"id": "2311.03732", "title": "Learning to Learn for Few-shot Continual Active Learning", "abstract": "  Continual learning strives to ensure stability in solving previously seen\ntasks while demonstrating plasticity in a novel domain. Recent advances in\ncontinual learning are mostly confined to a supervised learning setting,\nespecially in NLP domain. In this work, we consider a few-shot continual active\nlearning setting where labeled data are inadequate, and unlabeled data are\nabundant but with a limited annotation budget. We exploit meta-learning and\npropose a method, called Meta-Continual Active Learning. This method\nsequentially queries the most informative examples from a pool of unlabeled\ndata for annotation to enhance task-specific performance and tackle continual\nlearning problems through meta-objective. Specifically, we employ meta-learning\nand experience replay to address inter-task confusion and catastrophic\nforgetting. We further incorporate textual augmentations to avoid memory\nover-fitting caused by experience replay and sample queries, thereby ensuring\ngeneralization. We conduct extensive experiments on benchmark text\nclassification datasets from diverse domains to validate the feasibility and\neffectiveness of meta-continual active learning. We also analyze the impact of\ndifferent active learning strategies on various meta continual learning models.\nThe experimental results demonstrate that introducing randomness into sample\nselection is the best default strategy for maintaining generalization in\nmeta-continual learning framework.\n", "date": "2023-11-07", "categories": []}
{"id": "2404.08846", "title": "Experimental Design for Active Transductive Inference in Large Language\n  Models", "abstract": "  One emergent ability of large language models (LLMs) is that query-specific\nexamples can be included in the prompt at inference time. In this work, we use\nactive learning for adaptive prompt design and call it Active In-context Prompt\nDesign (AIPD). We design the LLM prompt by adaptively choosing few-shot\nexamples from a training set to optimize performance on a test set. The\ntraining examples are initially unlabeled and we obtain the label of the most\ninformative ones, which maximally reduces uncertainty in the LLM prediction. We\npropose two algorithms, GO and SAL, which differ in how the few-shot examples\nare chosen. We analyze these algorithms in linear models: first GO and then use\nits equivalence with SAL. We experiment with many different tasks in small,\nmedium-sized, and large language models; and show that GO and SAL outperform\nother methods for choosing few-shot examples in the LLM prompt at inference\ntime.\n", "date": "2024-04-12", "categories": []}
{"id": "2404.19043", "title": "Improving Interpretability of Deep Active Learning for Flood Inundation\n  Mapping Through Class Ambiguity Indices Using Multi-spectral Satellite\n  Imagery", "abstract": "  Flood inundation mapping is a critical task for responding to the increasing\nrisk of flooding linked to global warming. Significant advancements of deep\nlearning in recent years have triggered its extensive applications, including\nflood inundation mapping. To cope with the time-consuming and labor-intensive\ndata labeling process in supervised learning, deep active learning strategies\nare one of the feasible approaches. However, there remains limited exploration\ninto the interpretability of how deep active learning strategies operate, with\na specific focus on flood inundation mapping in the field of remote sensing. In\nthis study, we introduce a novel framework of Interpretable Deep Active\nLearning for Flood inundation Mapping (IDAL-FIM), specifically in terms of\nclass ambiguity of multi-spectral satellite images. In the experiments, we\nutilize Sen1Floods11 dataset, and adopt U-Net with MC-dropout. In addition, we\nemploy five acquisition functions, which are the random, K-means, BALD,\nentropy, and margin acquisition functions. Based on the experimental results,\nwe demonstrate that two proposed class ambiguity indices are effective\nvariables to interpret the deep active learning by establishing statistically\nsignificant correlation with the predictive uncertainty of the deep learning\nmodel at the tile level. Then, we illustrate the behaviors of deep active\nlearning through visualizing two-dimensional density plots and providing\ninterpretations regarding the operation of deep active learning, in flood\ninundation mapping.\n", "date": "2024-04-29", "categories": []}
{"id": "2405.17734", "title": "Towards Efficient Disaster Response via Cost-effective Unbiased Class\n  Rate Estimation through Neyman Allocation Stratified Sampling Active Learning", "abstract": "  With the rapid development of earth observation technology, we have entered\nan era of massively available satellite remote-sensing data. However, a large\namount of satellite remote sensing data lacks a label or the label cost is too\nhigh to hinder the potential of AI technology mining satellite data. Especially\nin such an emergency response scenario that uses satellite data to evaluate the\ndegree of disaster damage. Disaster damage assessment encountered bottlenecks\ndue to excessive focus on the damage of a certain building in a specific\ngeographical space or a certain area on a larger scale. In fact, in the early\ndays of disaster emergency response, government departments were more concerned\nabout the overall damage rate of the disaster area instead of single-building\ndamage, because this helps the government decide the level of emergency\nresponse. We present an innovative algorithm that constructs Neyman stratified\nrandom sampling trees for binary classification and extends this approach to\nmulticlass problems. Through extensive experimentation on various datasets and\nmodel structures, our findings demonstrate that our method surpasses both\npassive and conventional active learning techniques in terms of class rate\nestimation and model enhancement with only 30\\%-60\\% of the annotation cost of\nsimple sampling. It effectively addresses the 'sampling bias' challenge in\ntraditional active learning strategies and mitigates the 'cold start' dilemma.\nThe efficacy of our approach is further substantiated through application to\ndisaster evaluation tasks using Xview2 Satellite imagery, showcasing its\npractical utility in real-world contexts.\n", "date": "2024-05-28", "categories": []}
{"id": "2405.18078", "title": "Edge-guided and Class-balanced Active Learning for Semantic Segmentation\n  of Aerial Images", "abstract": "  Semantic segmentation requires pixel-level annotation, which is\ntime-consuming. Active Learning (AL) is a promising method for reducing data\nannotation costs. Due to the gap between aerial and natural images, the\nprevious AL methods are not ideal, mainly caused by unreasonable labeling units\nand the neglect of class imbalance. Previous labeling units are based on images\nor regions, which does not consider the characteristics of segmentation tasks\nand aerial images, i.e., the segmentation network often makes mistakes in the\nedge region, and the edge of aerial images is often interlaced and irregular.\nTherefore, an edge-guided labeling unit is proposed and supplemented as the new\nunit. On the other hand, the class imbalance is severe, manifested in two\naspects: the aerial image is seriously imbalanced, and the AL strategy does not\nfully consider the class balance. Both seriously affect the performance of AL\nin aerial images. We comprehensively ensure class balance from all steps that\nmay occur imbalance, including initial labeled data, subsequent labeled data,\nand pseudo-labels. Through the two improvements, our method achieves more than\n11.2\\% gains compared to state-of-the-art methods on three benchmark datasets,\nDeepglobe, Potsdam, and Vaihingen, and more than 18.6\\% gains compared to the\nbaseline. Sufficient ablation studies show that every module is indispensable.\nFurthermore, we establish a fair and strong benchmark for future research on AL\nfor aerial image segmentation.\n", "date": "2024-05-28", "categories": []}
{"id": "2309.12494", "title": "Evidential uncertainty sampling for active learning", "abstract": "  Recent studies in active learning, particularly in uncertainty sampling, have\nfocused on the decomposition of model uncertainty into reducible and\nirreducible uncertainties. In this paper, the aim is to simplify the\ncomputational process while eliminating the dependence on observations.\nCrucially, the inherent uncertainty in the labels is considered, the\nuncertainty of the oracles. Two strategies are proposed, sampling by Klir\nuncertainty, which tackles the exploration-exploitation dilemma, and sampling\nby evidential epistemic uncertainty, which extends the concept of reducible\nuncertainty within the evidential framework, both using the theory of belief\nfunctions. Experimental results in active learning demonstrate that our\nproposed method can outperform uncertainty sampling.\n", "date": "2023-09-21", "categories": []}
{"id": "2312.10361", "title": "Exploring UMAP in hybrid models of entropy-based and representativeness\n  sampling for active learning in biomedical segmentation", "abstract": "  In this work, we study various hybrid models of entropy-based and\nrepresentativeness sampling techniques in the context of active learning in\nmedical segmentation, in particular examining the role of UMAP (Uniform\nManifold Approximation and Projection) as a technique for capturing\nrepresentativeness. Although UMAP has been shown viable as a general purpose\ndimension reduction method in diverse areas, its role in deep learning-based\nmedical segmentation has yet been extensively explored. Using the cardiac and\nprostate datasets in the Medical Segmentation Decathlon for validation, we\nfound that a novel hybrid combination of Entropy-UMAP sampling technique\nachieved a statistically significant Dice score advantage over the random\nbaseline ($3.2 \\%$ for cardiac, $4.5 \\%$ for prostate), and attained the\nhighest Dice coefficient among the spectrum of 10 distinct active learning\nmethodologies we examined. This provides preliminary evidence that there is an\ninteresting synergy between entropy-based and UMAP methods when the former\nprecedes the latter in a hybrid model of active learning.\n", "date": "2023-12-16", "categories": []}
{"id": "2405.16301", "title": "Active Learning for Finely-Categorized Image-Text Retrieval by Selecting\n  Hard Negative Unpaired Samples", "abstract": "  Securing a sufficient amount of paired data is important to train an\nimage-text retrieval (ITR) model, but collecting paired data is very expensive.\nTo address this issue, in this paper, we propose an active learning algorithm\nfor ITR that can collect paired data cost-efficiently. Previous studies assume\nthat image-text pairs are given and their category labels are asked to the\nannotator. However, in the recent ITR studies, the importance of category label\nis decreased since a retrieval model can be trained with only image-text pairs.\nFor this reason, we set up an active learning scenario where unpaired images\n(or texts) are given and the annotator provides corresponding texts (or images)\nto make paired data. The key idea of the proposed AL algorithm is to select\nunpaired images (or texts) that can be hard negative samples for existing texts\n(or images). To this end, we introduce a novel scoring function to choose hard\nnegative samples. We validate the effectiveness of the proposed method on\nFlickr30K and MS-COCO datasets.\n", "date": "2024-05-25", "categories": []}
{"id": "2405.15031", "title": "Amortized nonmyopic active search via deep imitation learning", "abstract": "  Active search formalizes a specialized active learning setting where the goal\nis to collect members of a rare, valuable class. The state-of-the-art algorithm\napproximates the optimal Bayesian policy in a budget-aware manner, and has been\nshown to achieve impressive empirical performance in previous work. However,\neven this approximate policy has a superlinear computational complexity with\nrespect to the size of the search problem, rendering its application\nimpractical in large spaces or in real-time systems where decisions must be\nmade quickly. We study the amortization of this policy by training a neural\nnetwork to learn to search. To circumvent the difficulty of learning from\nscratch, we appeal to imitation learning techniques to mimic the behavior of\nthe expert, expensive-to-compute policy. Our policy network, trained on\nsynthetic data, learns a beneficial search strategy that yields nonmyopic\ndecisions carefully balancing exploration and exploitation. Extensive\nexperiments demonstrate our policy achieves competitive performance at\nreal-world tasks that closely approximates the expert's at a fraction of the\ncost, while outperforming cheaper baselines.\n", "date": "2024-05-23", "categories": []}
{"id": "2310.18786", "title": "A Competitive Algorithm for Agnostic Active Learning", "abstract": "  For some hypothesis classes and input distributions, active agnostic learning\nneeds exponentially fewer samples than passive learning; for other classes and\ndistributions, it offers little to no improvement. The most popular algorithms\nfor agnostic active learning express their performance in terms of a parameter\ncalled the disagreement coefficient, but it is known that these algorithms are\ninefficient on some inputs.\n  We take a different approach to agnostic active learning, getting an\nalgorithm that is competitive with the optimal algorithm for any binary\nhypothesis class $H$ and distribution $D_X$ over $X$. In particular, if any\nalgorithm can use $m^*$ queries to get $O(\\eta)$ error, then our algorithm uses\n$O(m^* \\log |H|)$ queries to get $O(\\eta)$ error. Our algorithm lies in the\nvein of the splitting-based approach of Dasgupta [2004], which gets a similar\nresult for the realizable ($\\eta = 0$) setting.\n  We also show that it is NP-hard to do better than our algorithm's $O(\\log\n|H|)$ overhead in general.\n", "date": "2023-10-28", "categories": []}
{"id": "2311.08053", "title": "Batch Selection and Communication for Active Learning with Edge Labeling", "abstract": "  Conventional retransmission (ARQ) protocols are designed with the goal of\nensuring the correct reception of all the individual transmitter's packets at\nthe receiver. When the transmitter is a learner communicating with a teacher,\nthis goal is at odds with the actual aim of the learner, which is that of\neliciting the most relevant label information from the teacher. Taking an\nactive learning perspective, this paper addresses the following key protocol\ndesign questions: (i) Active batch selection: Which batch of inputs should be\nsent to the teacher to acquire the most useful information and thus reduce the\nnumber of required communication rounds? (ii) Batch encoding: Can batches of\ndata points be combined to reduce the communication resources required at each\ncommunication round? Specifically, this work introduces\nCommunication-Constrained Bayesian Active Knowledge Distillation (CC-BAKD), a\nnovel protocol that integrates Bayesian active learning with compression via a\nlinear mix-up mechanism. Comparisons with existing active learning protocols\ndemonstrate the advantages of the proposed approach.\n", "date": "2023-11-14", "categories": []}
{"id": "2405.12122", "title": "An Active Learning Framework with a Class Balancing Strategy for Time\n  Series Classification", "abstract": "  Training machine learning models for classification tasks often requires\nlabeling numerous samples, which is costly and time-consuming, especially in\ntime series analysis. This research investigates Active Learning (AL)\nstrategies to reduce the amount of labeled data needed for effective time\nseries classification. Traditional AL techniques cannot control the selection\nof instances per class for labeling, leading to potential bias in\nclassification performance and instance selection, particularly in imbalanced\ntime series datasets. To address this, we propose a novel class-balancing\ninstance selection algorithm integrated with standard AL strategies. Our\napproach aims to select more instances from classes with fewer labeled\nexamples, thereby addressing imbalance in time series datasets. We demonstrate\nthe effectiveness of our AL framework in selecting informative data samples for\ntwo distinct domains of tactile texture recognition and industrial fault\ndetection. In robotics, our method achieves high-performance texture\ncategorization while significantly reducing labeled training data requirements\nto 70%. We also evaluate the impact of different sliding window time intervals\non robotic texture classification using AL strategies. In synthetic fiber\nmanufacturing, we adapt AL techniques to address the challenge of fault\nclassification, aiming to minimize data annotation cost and time for\nindustries. We also address real-life class imbalances in the multiclass\nindustrial anomalous dataset using our class-balancing instance algorithm\nintegrated with AL strategies. Overall, this thesis highlights the potential of\nour AL framework across these two distinct domains.\n", "date": "2024-05-20", "categories": []}
{"id": "2405.09817", "title": "Active Learning with Fully Bayesian Neural Networks for Discontinuous\n  and Nonstationary Data", "abstract": "  Active learning optimizes the exploration of large parameter spaces by\nstrategically selecting which experiments or simulations to conduct, thus\nreducing resource consumption and potentially accelerating scientific\ndiscovery. A key component of this approach is a probabilistic surrogate model,\ntypically a Gaussian Process (GP), which approximates an unknown functional\nrelationship between control parameters and a target property. However,\nconventional GPs often struggle when applied to systems with discontinuities\nand non-stationarities, prompting the exploration of alternative models. This\nlimitation becomes particularly relevant in physical science problems, which\nare often characterized by abrupt transitions between different system states\nand rapid changes in physical property behavior. Fully Bayesian Neural Networks\n(FBNNs) serve as a promising substitute, treating all neural network weights\nprobabilistically and leveraging advanced Markov Chain Monte Carlo techniques\nfor direct sampling from the posterior distribution. This approach enables\nFBNNs to provide reliable predictive distributions, crucial for making informed\ndecisions under uncertainty in the active learning setting. Although\ntraditionally considered too computationally expensive for 'big data'\napplications, many physical sciences problems involve small amounts of data in\nrelatively low-dimensional parameter spaces. Here, we assess the suitability\nand performance of FBNNs with the No-U-Turn Sampler for active learning tasks\nin the 'small data' regime, highlighting their potential to enhance predictive\naccuracy and reliability on test functions relevant to problems in physical\nsciences.\n", "date": "2024-05-16", "categories": []}
{"id": "2401.09787", "title": "Querying Easily Flip-flopped Samples for Deep Active Learning", "abstract": "  Active learning is a machine learning paradigm that aims to improve the\nperformance of a model by strategically selecting and querying unlabeled data.\nOne effective selection strategy is to base it on the model's predictive\nuncertainty, which can be interpreted as a measure of how informative a sample\nis. The sample's distance to the decision boundary is a natural measure of\npredictive uncertainty, but it is often intractable to compute, especially for\ncomplex decision boundaries formed in multiclass classification tasks. To\naddress this issue, this paper proposes the {\\it least disagree metric} (LDM),\ndefined as the smallest probability of disagreement of the predicted label, and\nan estimator for LDM proven to be asymptotically consistent under mild\nassumptions. The estimator is computationally efficient and can be easily\nimplemented for deep learning models using parameter perturbation. The\nLDM-based active learning is performed by querying unlabeled data with the\nsmallest LDM. Experimental results show that our LDM-based active learning\nalgorithm obtains state-of-the-art overall performance on all considered\ndatasets and deep architectures.\n", "date": "2024-01-18", "categories": []}
{"id": "2405.08921", "title": "Neural Active Learning Meets the Partial Monitoring Framework", "abstract": "  We focus on the online-based active learning (OAL) setting where an agent\noperates over a stream of observations and trades-off between the costly\nacquisition of information (labelled observations) and the cost of prediction\nerrors. We propose a novel foundation for OAL tasks based on partial\nmonitoring, a theoretical framework specialized in online learning from\npartially informative actions. We show that previously studied binary and\nmulti-class OAL tasks are instances of partial monitoring. We expand the\nreal-world potential of OAL by introducing a new class of cost-sensitive OAL\ntasks. We propose NeuralCBP, the first PM strategy that accounts for predictive\nuncertainty with deep neural networks. Our extensive empirical evaluation on\nopen source datasets shows that NeuralCBP has favorable performance against\nstate-of-the-art baselines on multiple binary, multi-class and cost-sensitive\nOAL tasks.\n", "date": "2024-05-14", "categories": []}
{"id": "2405.07816", "title": "Quick and Accurate Affordance Learning", "abstract": "  Infants learn actively in their environments, shaping their own learning\ncurricula. They learn about their environments' affordances, that is, how local\ncircumstances determine how their behavior can affect the environment. Here we\nmodel this type of behavior by means of a deep learning architecture. The\narchitecture mediates between global cognitive map exploration and local\naffordance learning. Inference processes actively move the simulated agent\ntowards regions where they expect affordance-related knowledge gain. We\ncontrast three measures of uncertainty to guide this exploration: predicted\nuncertainty of a model, standard deviation between the means of several models\n(SD), and the Jensen-Shannon Divergence (JSD) between several models. We show\nthat the first measure gets fooled by aleatoric uncertainty inherent in the\nenvironment, while the two other measures focus learning on epistemic\nuncertainty. JSD exhibits the most balanced exploration strategy. From a\ncomputational perspective, our model suggests three key ingredients for\ncoordinating the active generation of learning curricula: (1) Navigation\nbehavior needs to be coordinated with local motor behavior for enabling active\naffordance learning. (2) Affordances need to be encoded locally for acquiring\ngeneralized knowledge. (3) Effective active affordance learning mechanisms\nshould use density comparison techniques for estimating expected knowledge\ngain. Future work may seek collaborations with developmental psychology to\nmodel active play in children in more realistic scenarios.\n", "date": "2024-05-13", "categories": []}
{"id": "2405.07971", "title": "Sensitivity Analysis for Active Sampling, with Applications to the\n  Simulation of Analog Circuits", "abstract": "  We propose an active sampling flow, with the use-case of simulating the\nimpact of combined variations on analog circuits. In such a context, given the\nlarge number of parameters, it is difficult to fit a surrogate model and to\nefficiently explore the space of design features.\n  By combining a drastic dimension reduction using sensitivity analysis and\nBayesian surrogate modeling, we obtain a flexible active sampling flow. On\nsynthetic and real datasets, this flow outperforms the usual Monte-Carlo\nsampling which often forms the foundation of design space exploration.\n", "date": "2024-05-13", "categories": []}
{"id": "2402.10074", "title": "Class-Balanced and Reinforced Active Learning on Graphs", "abstract": "  Graph neural networks (GNNs) have demonstrated significant success in various\napplications, such as node classification, link prediction, and graph\nclassification. Active learning for GNNs aims to query the valuable samples\nfrom the unlabeled data for annotation to maximize the GNNs' performance at a\nlower cost. However, most existing algorithms for reinforced active learning in\nGNNs may lead to a highly imbalanced class distribution, especially in highly\nskewed class scenarios. GNNs trained with class-imbalanced labeled data are\nsusceptible to bias toward majority classes, and the lower performance of\nminority classes may lead to a decline in overall performance. To tackle this\nissue, we propose a novel class-balanced and reinforced active learning\nframework for GNNs, namely, GCBR. It learns an optimal policy to acquire\nclass-balanced and informative nodes for annotation, maximizing the performance\nof GNNs trained with selected labeled nodes. GCBR designs class-balance-aware\nstates, as well as a reward function that achieves trade-off between model\nperformance and class balance. The reinforcement learning algorithm Advantage\nActor-Critic (A2C) is employed to learn an optimal policy stably and\nefficiently. We further upgrade GCBR to GCBR++ by introducing a punishment\nmechanism to obtain a more class-balanced labeled set. Extensive experiments on\nmultiple datasets demonstrate the effectiveness of the proposed approaches,\nachieving superior performance over state-of-the-art baselines.\n", "date": "2024-02-15", "categories": []}
{"id": "2310.04966", "title": "Improved Active Learning via Dependent Leverage Score Sampling", "abstract": "  We show how to obtain improved active learning methods in the agnostic\n(adversarial noise) setting by combining marginal leverage score sampling with\nnon-independent sampling strategies that promote spatial coverage. In\nparticular, we propose an easily implemented method based on the \\emph{pivotal\nsampling algorithm}, which we test on problems motivated by learning-based\nmethods for parametric PDEs and uncertainty quantification. In comparison to\nindependent sampling, our method reduces the number of samples needed to reach\na given target accuracy by up to $50\\%$. We support our findings with two\ntheoretical results. First, we show that any non-independent leverage score\nsampling method that obeys a weak \\emph{one-sided $\\ell_{\\infty}$ independence\ncondition} (which includes pivotal sampling) can actively learn $d$ dimensional\nlinear functions with $O(d\\log d)$ samples, matching independent sampling. This\nresult extends recent work on matrix Chernoff bounds under $\\ell_{\\infty}$\nindependence, and may be of interest for analyzing other sampling strategies\nbeyond pivotal sampling. Second, we show that, for the important case of\npolynomial regression, our pivotal method obtains an improved bound on $O(d)$\nsamples.\n", "date": "2023-10-08", "categories": []}
{"id": "2405.01616", "title": "Generative Active Learning for the Search of Small-molecule Protein\n  Binders", "abstract": "  Despite substantial progress in machine learning for scientific discovery in\nrecent years, truly de novo design of small molecules which exhibit a property\nof interest remains a significant challenge. We introduce LambdaZero, a\ngenerative active learning approach to search for synthesizable molecules.\nPowered by deep reinforcement learning, LambdaZero learns to search over the\nvast space of molecules to discover candidates with a desired property. We\napply LambdaZero with molecular docking to design novel small molecules that\ninhibit the enzyme soluble Epoxide Hydrolase 2 (sEH), while enforcing\nconstraints on synthesizability and drug-likeliness. LambdaZero provides an\nexponential speedup in terms of the number of calls to the expensive molecular\ndocking oracle, and LambdaZero de novo designed molecules reach docking scores\nthat would otherwise require the virtual screening of a hundred billion\nmolecules. Importantly, LambdaZero discovers novel scaffolds of synthesizable,\ndrug-like inhibitors for sEH. In in vitro experimental validation, a series of\nligands from a generated quinazoline-based scaffold were synthesized, and the\nlead inhibitor N-(4,6-di(pyrrolidin-1-yl)quinazolin-2-yl)-N-methylbenzamide\n(UM0152893) displayed sub-micromolar enzyme inhibition of sEH.\n", "date": "2024-05-02", "categories": []}
{"id": "2404.17249", "title": "Making Better Use of Unlabelled Data in Bayesian Active Learning", "abstract": "  Fully supervised models are predominant in Bayesian active learning. We argue\nthat their neglect of the information present in unlabelled data harms not just\npredictive performance but also decisions about what data to acquire. Our\nproposed solution is a simple framework for semi-supervised Bayesian active\nlearning. We find it produces better-performing models than either conventional\nBayesian active learning or semi-supervised learning with randomly acquired\ndata. It is also easier to scale up than the conventional approach. As well as\nsupporting a shift towards semi-supervised models, our findings highlight the\nimportance of studying models and acquisition methods in conjunction.\n", "date": "2024-04-26", "categories": []}
{"id": "2208.03346", "title": "Active Learning for Non-Parametric Choice Models", "abstract": "  We study the problem of actively learning a non-parametric choice model based\non consumers' decisions. We present a negative result showing that such choice\nmodels may not be identifiable. To overcome the identifiability problem, we\nintroduce a directed acyclic graph (DAG) representation of the choice model.\nThis representation provably encodes all the information about the choice model\nwhich can be inferred from the available data, in the sense that it permits\ncomputing all choice probabilities.\n  We establish that given exact choice probabilities for a collection of item\nsets, one can reconstruct the DAG. However, attempting to extend this\nmethodology to estimate the DAG from noisy choice frequency data obtained\nduring an active learning process leads to inaccuracies. To address this\nchallenge, we present an inclusion-exclusion approach that effectively manages\nerror propagation across DAG levels, leading to a more accurate estimate of the\nDAG. Utilizing this technique, our algorithm estimates the DAG representation\nof an underlying non-parametric choice model. The algorithm operates\nefficiently (in polynomial time) when the set of frequent rankings is drawn\nuniformly at random. It learns the distribution over the most popular items\namong frequent preference types by actively and repeatedly offering assortments\nof items and observing the chosen item. We demonstrate that our algorithm more\neffectively recovers a set of frequent preferences on both synthetic and\npublicly available datasets on consumers' preferences, compared to\ncorresponding non-active learning estimation algorithms. These findings\nunderscore the value of our algorithm and the broader applicability of\nactive-learning approaches in modeling consumer behavior.\n", "date": "2022-08-05", "categories": []}
{"id": "2310.16161", "title": "MyriadAL: Active Few Shot Learning for Histopathology", "abstract": "  Active Learning (AL) and Few Shot Learning (FSL) are two label-efficient\nmethods which have achieved excellent results recently. However, most prior\narts in both learning paradigms fail to explore the wealth of the vast\nunlabelled data. In this study, we address this issue in the scenario where the\nannotation budget is very limited, yet a large amount of unlabelled data for\nthe target task is available. We frame this work in the context of\nhistopathology where labelling is prohibitively expensive. To this end, we\nintroduce an active few shot learning framework, Myriad Active Learning (MAL),\nincluding a contrastive-learning encoder, pseudo-label generation, and novel\nquery sample selection in the loop. Specifically, we propose to massage\nunlabelled data in a self-supervised manner, where the obtained data\nrepresentations and clustering knowledge form the basis to activate the AL\nloop. With feedback from the oracle in each AL cycle, the pseudo-labels of the\nunlabelled data are refined by optimizing a shallow task-specific net on top of\nthe encoder. These updated pseudo-labels serve to inform and improve the active\nlearning query selection process. Furthermore, we introduce a novel recipe to\ncombine existing uncertainty measures and utilize the entire uncertainty list\nto reduce sample redundancy in AL. Extensive experiments on two public\nhistopathology datasets show that MAL has superior test accuracy, macro\nF1-score, and label efficiency compared to prior works, and can achieve a\ncomparable test accuracy to a fully supervised algorithm while labelling only\n5% of the dataset.\n", "date": "2023-10-24", "categories": []}
{"id": "2312.02567", "title": "Think Twice Before Selection: Federated Evidential Active Learning for\n  Medical Image Analysis with Domain Shifts", "abstract": "  Federated learning facilitates the collaborative learning of a global model\nacross multiple distributed medical institutions without centralizing data.\nNevertheless, the expensive cost of annotation on local clients remains an\nobstacle to effectively utilizing local data. To mitigate this issue, federated\nactive learning methods suggest leveraging local and global model predictions\nto select a relatively small amount of informative local data for annotation.\nHowever, existing methods mainly focus on all local data sampled from the same\ndomain, making them unreliable in realistic medical scenarios with domain\nshifts among different clients. In this paper, we make the first attempt to\nassess the informativeness of local data derived from diverse domains and\npropose a novel methodology termed Federated Evidential Active Learning (FEAL)\nto calibrate the data evaluation under domain shift. Specifically, we introduce\na Dirichlet prior distribution in both local and global models to treat the\nprediction as a distribution over the probability simplex and capture both\naleatoric and epistemic uncertainties by using the Dirichlet-based evidential\nmodel. Then we employ the epistemic uncertainty to calibrate the aleatoric\nuncertainty. Afterward, we design a diversity relaxation strategy to reduce\ndata redundancy and maintain data diversity. Extensive experiments and analysis\non five real multi-center medical image datasets demonstrate the superiority of\nFEAL over the state-of-the-art active learning methods in federated scenarios\nwith domain shifts. The code will be available at\nhttps://github.com/JiayiChen815/FEAL.\n", "date": "2023-12-05", "categories": []}
{"id": "2404.12522", "title": "Neural Active Learning Beyond Bandits", "abstract": "  We study both stream-based and pool-based active learning with neural network\napproximations. A recent line of works proposed bandit-based approaches that\ntransformed active learning into a bandit problem, achieving both theoretical\nand empirical success. However, the performance and computational costs of\nthese methods may be susceptible to the number of classes, denoted as $K$, due\nto this transformation. Therefore, this paper seeks to answer the question:\n\"How can we mitigate the adverse impacts of $K$ while retaining the advantages\nof principled exploration and provable performance guarantees in active\nlearning?\" To tackle this challenge, we propose two algorithms based on the\nnewly designed exploitation and exploration neural networks for stream-based\nand pool-based active learning. Subsequently, we provide theoretical\nperformance guarantees for both algorithms in a non-parametric setting,\ndemonstrating a slower error-growth rate concerning $K$ for the proposed\napproaches. We use extensive experiments to evaluate the proposed algorithms,\nwhich consistently outperform state-of-the-art baselines.\n", "date": "2024-04-18", "categories": []}
{"id": "2306.13935", "title": "Are Good Explainers Secretly Human-in-the-Loop Active Learners?", "abstract": "  Explainable AI (XAI) techniques have become popular for multiple use-cases in\nthe past few years. Here we consider its use in studying model predictions to\ngather additional training data. We argue that this is equivalent to Active\nLearning, where the query strategy involves a human-in-the-loop. We provide a\nmathematical approximation for the role of the human, and present a general\nformalization of the end-to-end workflow. This enables us to rigorously compare\nthis use with standard Active Learning algorithms, while allowing for\nextensions to the workflow. An added benefit is that their utility can be\nassessed via simulation instead of conducting expensive user-studies. We also\npresent some initial promising results.\n", "date": "2023-06-24", "categories": []}
{"id": "2401.07853", "title": "VeCAF: Vision-language Collaborative Active Finetuning with Training\n  Objective Awareness", "abstract": "  Finetuning a pretrained vision model (PVM) is a common technique for learning\ndownstream vision tasks. However, the conventional finetuning process with\nrandomly sampled data points results in diminished training efficiency. To\naddress this drawback, we propose a novel approach, Vision-language\nCollaborative Active Finetuning (VeCAF). With the emerging availability of\nlabels and natural language annotations of images through web-scale crawling or\ncontrolled generation, VeCAF makes use of these information to perform\nparametric data selection for PVM finetuning. VeCAF incorporates the finetuning\nobjective to select significant data points that effectively guide the PVM\ntowards faster convergence to meet the performance goal. This process is\nassisted by the inherent semantic richness of the text embedding space which we\nuse to augment image features. Furthermore, the flexibility of text-domain\naugmentation allows VeCAF to handle out-of-distribution scenarios without\nexternal data. Extensive experiments show the leading performance and high\ncomputational efficiency of VeCAF that is superior to baselines in both\nin-distribution and out-of-distribution image classification tasks. On\nImageNet, VeCAF uses up to 3.3x less training batches to reach the target\nperformance compared to full finetuning, and achieves an accuracy improvement\nof 2.7% over the state-of-the-art active finetuning method with the same number\nof batches.\n", "date": "2024-01-15", "categories": []}
{"id": "2402.18260", "title": "Efficiently Computable Safety Bounds for Gaussian Processes in Active\n  Learning", "abstract": "  Active learning of physical systems must commonly respect practical safety\nconstraints, which restricts the exploration of the design space. Gaussian\nProcesses (GPs) and their calibrated uncertainty estimations are widely used\nfor this purpose. In many technical applications the design space is explored\nvia continuous trajectories, along which the safety needs to be assessed. This\nis particularly challenging for strict safety requirements in GP methods, as it\nemploys computationally expensive Monte-Carlo sampling of high quantiles. We\naddress these challenges by providing provable safety bounds based on the\nadaptively sampled median of the supremum of the posterior GP. Our method\nsignificantly reduces the number of samples required for estimating high safety\nprobabilities, resulting in faster evaluation without sacrificing accuracy and\nexploration speed. The effectiveness of our safe active learning approach is\ndemonstrated through extensive simulations and validated using a real-world\nengine example.\n", "date": "2024-02-28", "categories": []}
{"id": "2404.09953", "title": "Classification Tree-based Active Learning: A Wrapper Approach", "abstract": "  Supervised machine learning often requires large training sets to train\naccurate models, yet obtaining large amounts of labeled data is not always\nfeasible. Hence, it becomes crucial to explore active learning methods for\nreducing the size of training sets while maintaining high accuracy. The aim is\nto select the optimal subset of data for labeling from an initial unlabeled\nset, ensuring precise prediction of outcomes. However, conventional active\nlearning approaches are comparable to classical random sampling. This paper\nproposes a wrapper active learning method for classification, organizing the\nsampling process into a tree structure, that improves state-of-the-art\nalgorithms. A classification tree constructed on an initial set of labeled\nsamples is considered to decompose the space into low-entropy regions.\nInput-space based criteria are used thereafter to sub-sample from these\nregions, the total number of points to be labeled being decomposed into each\nregion. This adaptation proves to be a significant enhancement over existing\nactive learning methods. Through experiments conducted on various benchmark\ndata sets, the paper demonstrates the efficacy of the proposed framework by\nbeing effective in constructing accurate classification models, even when\nprovided with a severely restricted labeled data set.\n", "date": "2024-04-15", "categories": []}
{"id": "2404.08078", "title": "SQBC: Active Learning using LLM-Generated Synthetic Data for Stance\n  Detection in Online Political Discussions", "abstract": "  Stance detection is an important task for many applications that analyse or\nsupport online political discussions. Common approaches include fine-tuning\ntransformer based models. However, these models require a large amount of\nlabelled data, which might not be available. In this work, we present two\ndifferent ways to leverage LLM-generated synthetic data to train and improve\nstance detection agents for online political discussions: first, we show that\naugmenting a small fine-tuning dataset with synthetic data can improve the\nperformance of the stance detection model. Second, we propose a new active\nlearning method called SQBC based on the \"Query-by-Comittee\" approach. The key\nidea is to use LLM-generated synthetic data as an oracle to identify the most\ninformative unlabelled samples, that are selected for manual labelling.\nComprehensive experiments show that both ideas can improve the stance detection\nperformance. Curiously, we observed that fine-tuning on actively selected\nsamples can exceed the performance of using the full dataset.\n", "date": "2024-04-11", "categories": []}
{"id": "2404.07663", "title": "Interactive Ontology Matching with Cost-Efficient Learning", "abstract": "  The creation of high-quality ontologies is crucial for data integration and\nknowledge-based reasoning, specifically in the context of the rising data\neconomy. However, automatic ontology matchers are often bound to the heuristics\nthey are based on, leaving many matches unidentified. Interactive ontology\nmatching systems involving human experts have been introduced, but they do not\nsolve the fundamental issue of flexibly finding additional matches outside the\nscope of the implemented heuristics, even though this is highly demanded in\nindustrial settings. Active machine learning methods appear to be a promising\npath towards a flexible interactive ontology matcher. However, off-the-shelf\nactive learning mechanisms suffer from low query efficiency due to extreme\nclass imbalance, resulting in a last-mile problem where high human effort is\nrequired to identify the remaining matches.\n  To address the last-mile problem, this work introduces DualLoop, an active\nlearning method tailored to ontology matching. DualLoop offers three main\ncontributions: (1) an ensemble of tunable heuristic matchers, (2) a short-term\nlearner with a novel query strategy adapted to highly imbalanced data, and (3)\nlong-term learners to explore potential matches by creating and tuning new\nheuristics. We evaluated DualLoop on three datasets of varying sizes and\ndomains. Compared to existing active learning methods, we consistently achieved\nbetter F1 scores and recall, reducing the expected query cost spent on finding\n90% of all matches by over 50%. Compared to traditional interactive ontology\nmatchers, we are able to find additional, last-mile matches. Finally, we detail\nthe successful deployment of our approach within an actual product and report\nits operational performance results within the Architecture, Engineering, and\nConstruction (AEC) industry sector, showcasing its practical value and\nefficiency.\n", "date": "2024-04-11", "categories": []}
{"id": "2404.04663", "title": "Focused Active Learning for Histopathological Image Classification", "abstract": "  Active Learning (AL) has the potential to solve a major problem of digital\npathology: the efficient acquisition of labeled data for machine learning\nalgorithms. However, existing AL methods often struggle in realistic settings\nwith artifacts, ambiguities, and class imbalances, as commonly seen in the\nmedical field. The lack of precise uncertainty estimations leads to the\nacquisition of images with a low informative value. To address these\nchallenges, we propose Focused Active Learning (FocAL), which combines a\nBayesian Neural Network with Out-of-Distribution detection to estimate\ndifferent uncertainties for the acquisition function. Specifically, the\nweighted epistemic uncertainty accounts for the class imbalance, aleatoric\nuncertainty for ambiguous images, and an OoD score for artifacts. We perform\nextensive experiments to validate our method on MNIST and the real-world Panda\ndataset for the classification of prostate cancer. The results confirm that\nother AL methods are 'distracted' by ambiguities and artifacts which harm the\nperformance. FocAL effectively focuses on the most informative images, avoiding\nambiguities and artifacts during acquisition. For both experiments, FocAL\noutperforms existing AL approaches, reaching a Cohen's kappa of 0.764 with only\n0.69% of the labeled Panda data.\n", "date": "2024-04-06", "categories": []}
{"id": "2404.05094", "title": "Active Test-Time Adaptation: Theoretical Analyses and An Algorithm", "abstract": "  Test-time adaptation (TTA) addresses distribution shifts for streaming test\ndata in unsupervised settings. Currently, most TTA methods can only deal with\nminor shifts and rely heavily on heuristic and empirical studies.\n  To advance TTA under domain shifts, we propose the novel problem setting of\nactive test-time adaptation (ATTA) that integrates active learning within the\nfully TTA setting.\n  We provide a learning theory analysis, demonstrating that incorporating\nlimited labeled test instances enhances overall performances across test\ndomains with a theoretical guarantee. We also present a sample entropy\nbalancing for implementing ATTA while avoiding catastrophic forgetting (CF). We\nintroduce a simple yet effective ATTA algorithm, known as SimATTA, using\nreal-time sample selection techniques. Extensive experimental results confirm\nconsistency with our theoretical analyses and show that the proposed ATTA\nmethod yields substantial performance improvements over TTA methods while\nmaintaining efficiency and shares similar effectiveness to the more demanding\nactive domain adaptation (ADA) methods. Our code is available at\nhttps://github.com/divelab/ATTA\n", "date": "2024-04-07", "categories": []}
{"id": "2404.04224", "title": "Active Causal Learning for Decoding Chemical Complexities with Targeted\n  Interventions", "abstract": "  Predicting and enhancing inherent properties based on molecular structures is\nparamount to design tasks in medicine, materials science, and environmental\nmanagement. Most of the current machine learning and deep learning approaches\nhave become standard for predictions, but they face challenges when applied\nacross different datasets due to reliance on correlations between molecular\nrepresentation and target properties. These approaches typically depend on\nlarge datasets to capture the diversity within the chemical space, facilitating\na more accurate approximation, interpolation, or extrapolation of the chemical\nbehavior of molecules. In our research, we introduce an active learning\napproach that discerns underlying cause-effect relationships through strategic\nsampling with the use of a graph loss function. This method identifies the\nsmallest subset of the dataset capable of encoding the most information\nrepresentative of a much larger chemical space. The identified causal relations\nare then leveraged to conduct systematic interventions, optimizing the design\ntask within a chemical space that the models have not encountered previously.\nWhile our implementation focused on the QM9 quantum-chemical dataset for a\nspecific design task-finding molecules with a large dipole moment-our active\ncausal learning approach, driven by intelligent sampling and interventions,\nholds potential for broader applications in molecular, materials design and\ndiscovery.\n", "date": "2024-04-05", "categories": []}
{"id": "2404.02580", "title": "Active learning for efficient annotation in precision agriculture: a\n  use-case on crop-weed semantic segmentation", "abstract": "  Optimizing deep learning models requires large amounts of annotated images, a\nprocess that is both time-intensive and costly. Especially for semantic\nsegmentation models in which every pixel must be annotated. A potential\nstrategy to mitigate annotation effort is active learning. Active learning\nfacilitates the identification and selection of the most informative images\nfrom a large unlabelled pool. The underlying premise is that these selected\nimages can improve the model's performance faster than random selection to\nreduce annotation effort. While active learning has demonstrated promising\nresults on benchmark datasets like Cityscapes, its performance in the\nagricultural domain remains largely unexplored. This study addresses this\nresearch gap by conducting a comparative study of three active learning-based\nacquisition functions: Bayesian Active Learning by Disagreement (BALD),\nstochastic-based BALD (PowerBALD), and Random. The acquisition functions were\ntested on two agricultural datasets: Sugarbeet and Corn-Weed, both containing\nthree semantic classes: background, crop and weed. Our results indicated that\nactive learning, especially PowerBALD, yields a higher performance than Random\nsampling on both datasets. But due to the relatively large standard deviations,\nthe differences observed were minimal; this was partly caused by high image\nredundancy and imbalanced classes. Specifically, more than 89\\% of the pixels\nbelonged to the background class on both datasets. The absence of significant\nresults on both datasets indicates that further research is required for\napplying active learning on agricultural datasets, especially if they contain a\nhigh-class imbalance and redundant images. Recommendations and insights are\nprovided in this paper to potentially resolve such issues.\n", "date": "2024-04-03", "categories": []}
{"id": "2404.01588", "title": "Hallucination Diversity-Aware Active Learning for Text Summarization", "abstract": "  Large Language Models (LLMs) have shown propensity to generate hallucinated\noutputs, i.e., texts that are factually incorrect or unsupported. Existing\nmethods for alleviating hallucinations typically require costly human\nannotations to identify and correct hallucinations in LLM outputs. Moreover,\nmost of these methods focus on a specific type of hallucination, e.g., entity\nor token errors, which limits their effectiveness in addressing various types\nof hallucinations exhibited in LLM outputs. To our best knowledge, in this\npaper we propose the first active learning framework to alleviate LLM\nhallucinations, reducing costly human annotations of hallucination needed. By\nmeasuring fine-grained hallucinations from errors in semantic frame, discourse\nand content verifiability in text summarization, we propose HAllucination\nDiversity-Aware Sampling (HADAS) to select diverse hallucinations for\nannotations in active learning for LLM finetuning. Extensive experiments on\nthree datasets and different backbone models demonstrate advantages of our\nmethod in effectively and efficiently mitigating LLM hallucinations.\n", "date": "2024-04-02", "categories": []}
{"id": "2309.16971", "title": "Multi-Resolution Active Learning of Fourier Neural Operators", "abstract": "  Fourier Neural Operator (FNO) is a popular operator learning framework. It\nnot only achieves the state-of-the-art performance in many tasks, but also is\nefficient in training and prediction. However, collecting training data for the\nFNO can be a costly bottleneck in practice, because it often demands expensive\nphysical simulations. To overcome this problem, we propose Multi-Resolution\nActive learning of FNO (MRA-FNO), which can dynamically select the input\nfunctions and resolutions to lower the data cost as much as possible while\noptimizing the learning efficiency. Specifically, we propose a probabilistic\nmulti-resolution FNO and use ensemble Monte-Carlo to develop an effective\nposterior inference algorithm. To conduct active learning, we maximize a\nutility-cost ratio as the acquisition function to acquire new examples and\nresolutions at each step. We use moment matching and the matrix determinant\nlemma to enable tractable, efficient utility computation. Furthermore, we\ndevelop a cost annealing framework to avoid over-penalizing high-resolution\nqueries at the early stage. The over-penalization is severe when the cost\ndifference is significant between the resolutions, which renders active\nlearning often stuck at low-resolution queries and inferior performance. Our\nmethod overcomes this problem and applies to general multi-fidelity active\nlearning and optimization problems. We have shown the advantage of our method\nin several benchmark operator learning tasks. The code is available at\nhttps://github.com/shib0li/MRA-FNO.\n", "date": "2023-09-29", "categories": []}
{"id": "2402.02561", "title": "Foundation Model Makes Clustering A Better Initialization For Cold-Start\n  Active Learning", "abstract": "  Active learning selects the most informative samples from the unlabelled\ndataset to annotate in the context of a limited annotation budget. While\nnumerous methods have been proposed for subsequent sample selection based on an\ninitialized model, scant attention has been paid to the indispensable phase of\nactive learning: selecting samples for model cold-start initialization. Most of\nthe previous studies resort to random sampling or naive clustering. However,\nrandom sampling is prone to fluctuation, and naive clustering suffers from\nconvergence speed, particularly when dealing with high-dimensional data such as\nimaging data. In this work, we propose to integrate foundation models with\nclustering methods to select samples for cold-start active learning\ninitialization. Foundation models refer to those trained on massive datasets by\nthe self-supervised paradigm and capable of generating informative and\ncompacted embeddings for various downstream tasks. Leveraging these embeddings\nto replace raw features such as pixel values, clustering quickly converges and\nidentifies better initial samples. For a comprehensive comparison, we included\na classic ImageNet-supervised model to acquire embeddings. Experiments on two\nclinical tasks of image classification and segmentation demonstrated that\nfoundation model-based clustering efficiently pinpointed informative initial\nsamples, leading to models showcasing enhanced performance than the baseline\nmethods. We envisage that this study provides an effective paradigm for future\ncold-start active learning.\n", "date": "2024-02-04", "categories": []}
{"id": "2403.18436", "title": "Collaborative Active Learning in Conditional Trust Environment", "abstract": "  In this paper, we investigate collaborative active learning, a paradigm in\nwhich multiple collaborators explore a new domain by leveraging their combined\nmachine learning capabilities without disclosing their existing data and\nmodels. Instead, the collaborators share prediction results from the new domain\nand newly acquired labels. This collaboration offers several advantages: (a) it\naddresses privacy and security concerns by eliminating the need for direct\nmodel and data disclosure; (b) it enables the use of different data sources and\ninsights without direct data exchange; and (c) it promotes cost-effectiveness\nand resource efficiency through shared labeling costs. To realize these\nbenefits, we introduce a collaborative active learning framework designed to\nfulfill the aforementioned objectives. We validate the effectiveness of the\nproposed framework through simulations. The results demonstrate that\ncollaboration leads to higher AUC scores compared to independent efforts,\nhighlighting the framework's ability to overcome the limitations of individual\nmodels. These findings support the use of collaborative approaches in active\nlearning, emphasizing their potential to enhance outcomes through collective\nexpertise and shared resources. Our work provides a foundation for further\nresearch on collaborative active learning and its practical applications in\nvarious domains where data privacy, cost efficiency, and model performance are\ncritical considerations.\n", "date": "2024-03-27", "categories": []}
{"id": "2403.17233", "title": "Active Learning of Dynamics Using Prior Domain Knowledge in the Sampling\n  Process", "abstract": "  We present an active learning algorithm for learning dynamics that leverages\nside information by explicitly incorporating prior domain knowledge into the\nsampling process. Our proposed algorithm guides the exploration toward regions\nthat demonstrate high empirical discrepancy between the observed data and an\nimperfect prior model of the dynamics derived from side information. Through\nnumerical experiments, we demonstrate that this strategy explores regions of\nhigh discrepancy and accelerates learning while simultaneously reducing model\nuncertainty. We rigorously prove that our active learning algorithm yields a\nconsistent estimate of the underlying dynamics by providing an explicit rate of\nconvergence for the maximum predictive variance. We demonstrate the efficacy of\nour approach on an under-actuated pendulum system and on the half-cheetah\nMuJoCo environment.\n", "date": "2024-03-25", "categories": []}
{"id": "2403.14800", "title": "Deep Active Learning: A Reality Check", "abstract": "  We conduct a comprehensive evaluation of state-of-the-art deep active\nlearning methods. Surprisingly, under general settings, no single-model method\ndecisively outperforms entropy-based active learning, and some even fall short\nof random sampling. We delve into overlooked aspects like starting budget,\nbudget step, and pretraining's impact, revealing their significance in\nachieving superior results. Additionally, we extend our evaluation to other\ntasks, exploring the active learning effectiveness in combination with\nsemi-supervised learning, and object detection. Our experiments provide\nvaluable insights and concrete recommendations for future active learning\nstudies. By uncovering the limitations of current methods and understanding the\nimpact of different experimental settings, we aim to inspire more efficient\ntraining of deep learning models in real-world scenarios with limited\nannotation budgets. This work contributes to advancing active learning's\nefficacy in deep learning and empowers researchers to make informed decisions\nwhen applying active learning to their tasks.\n", "date": "2024-03-21", "categories": []}
{"id": "2403.15108", "title": "Active Learning for Regression based on Wasserstein distance and\n  GroupSort Neural Networks", "abstract": "  This paper addresses a new active learning strategy for regression problems.\nThe presented Wasserstein active regression model is based on the principles of\ndistribution-matching to measure the representativeness of the labeled dataset.\nThe Wasserstein distance is computed using GroupSort Neural Networks. The use\nof such networks provides theoretical foundations giving a way to quantify\nerrors with explicit bounds for their size and depth. This solution is combined\nwith another uncertainty-based approach that is more outlier-tolerant to\ncomplete the query strategy. Finally, this method is compared with other\nclassical and recent solutions. The study empirically shows the pertinence of\nsuch a representativity-uncertainty approach, which provides good estimation\nall along the query procedure. Moreover, the Wasserstein active regression\noften achieves more precise estimations and tends to improve accuracy faster\nthan other models.\n", "date": "2024-03-22", "categories": []}
{"id": "2311.11178", "title": "Active Prompt Learning in Vision Language Models", "abstract": "  Pre-trained Vision Language Models (VLMs) have demonstrated notable progress\nin various zero-shot tasks, such as classification and retrieval. Despite their\nperformance, because improving performance on new tasks requires task-specific\nknowledge, their adaptation is essential. While labels are needed for the\nadaptation, acquiring them is typically expensive. To overcome this challenge,\nactive learning, a method of achieving a high performance by obtaining labels\nfor a small number of samples from experts, has been studied. Active learning\nprimarily focuses on selecting unlabeled samples for labeling and leveraging\nthem to train models. In this study, we pose the question, \"how can the\npre-trained VLMs be adapted under the active learning framework?\" In response\nto this inquiry, we observe that (1) simply applying a conventional active\nlearning framework to pre-trained VLMs even may degrade performance compared to\nrandom selection because of the class imbalance in labeling candidates, and (2)\nthe knowledge of VLMs can provide hints for achieving the balance before\nlabeling. Based on these observations, we devise a novel active learning\nframework for VLMs, denoted as PCB. To assess the effectiveness of our\napproach, we conduct experiments on seven different real-world datasets, and\nthe results demonstrate that PCB surpasses conventional active learning and\nrandom sampling methods. Code will be available in\nhttps://github.com/kaist-dmlab/pcb .\n", "date": "2023-11-18", "categories": []}
{"id": "2403.14002", "title": "Uncertainty Driven Active Learning for Image Segmentation in Underwater\n  Inspection", "abstract": "  Active learning aims to select the minimum amount of data to train a model\nthat performs similarly to a model trained with the entire dataset. We study\nthe potential of active learning for image segmentation in underwater\ninfrastructure inspection tasks, where large amounts of data are typically\ncollected. The pipeline inspection images are usually semantically repetitive\nbut with great variations in quality. We use mutual information as the\nacquisition function, calculated using Monte Carlo dropout. To assess the\neffectiveness of the framework, DenseNet and HyperSeg are trained with the\nCamVid dataset using active learning. In addition, HyperSeg is trained with a\npipeline inspection dataset of over 50,000 images. For the pipeline dataset,\nHyperSeg with active learning achieved 67.5% meanIoU using 12.5% of the data,\nand 61.4% with the same amount of randomly selected images. This shows that\nusing active learning for segmentation models in underwater inspection tasks\ncan lower the cost significantly.\n", "date": "2024-03-20", "categories": []}
{"id": "2403.14350", "title": "Annotation-Efficient Polyp Segmentation via Active Learning", "abstract": "  Deep learning-based techniques have proven effective in polyp segmentation\ntasks when provided with sufficient pixel-wise labeled data. However, the high\ncost of manual annotation has created a bottleneck for model generalization. To\nminimize annotation costs, we propose a deep active learning framework for\nannotation-efficient polyp segmentation. In practice, we measure the\nuncertainty of each sample by examining the similarity between features masked\nby the prediction map of the polyp and the background area. Since the\nsegmentation model tends to perform weak in samples with indistinguishable\nfeatures of foreground and background areas, uncertainty sampling facilitates\nthe fitting of under-learning data. Furthermore, clustering image-level\nfeatures weighted by uncertainty identify samples that are both uncertain and\nrepresentative. To enhance the selectivity of the active selection strategy, we\npropose a novel unsupervised feature discrepancy learning mechanism. The\nselection strategy and feature optimization work in tandem to achieve optimal\nperformance with a limited annotation budget. Extensive experimental results\nhave demonstrated that our proposed method achieved state-of-the-art\nperformance compared to other competitors on both a public dataset and a\nlarge-scale in-house dataset.\n", "date": "2024-03-21", "categories": []}
{"id": "2307.04536", "title": "DADO -- Low-Cost Query Strategies for Deep Active Design Optimization", "abstract": "  In this experience report, we apply deep active learning to the field of\ndesign optimization to reduce the number of computationally expensive numerical\nsimulations. We are interested in optimizing the design of structural\ncomponents, where the shape is described by a set of parameters. If we can\npredict the performance based on these parameters and consider only the\npromising candidates for simulation, there is an enormous potential for saving\ncomputing power. We present two selection strategies for self-optimization to\nreduce the computational cost in multi-objective design optimization problems.\nOur proposed methodology provides an intuitive approach that is easy to apply,\noffers significant improvements over random sampling, and circumvents the need\nfor uncertainty estimation. We evaluate our strategies on a large dataset from\nthe domain of fluid dynamics and introduce two new evaluation metrics to\ndetermine the model's performance. Findings from our evaluation highlights the\neffectiveness of our selection strategies in accelerating design optimization.\nWe believe that the introduced method is easily transferable to other\nself-optimization problems.\n", "date": "2023-07-10", "categories": []}
{"id": "2403.10069", "title": "Boundary Matters: A Bi-Level Active Finetuning Framework", "abstract": "  The pretraining-finetuning paradigm has gained widespread adoption in vision\ntasks and other fields, yet it faces the significant challenge of high sample\nannotation costs. To mitigate this, the concept of active finetuning has\nemerged, aiming to select the most appropriate samples for model finetuning\nwithin a limited budget. Traditional active learning methods often struggle in\nthis setting due to their inherent bias in batch selection. Furthermore, the\nrecent active finetuning approach has primarily concentrated on aligning the\ndistribution of selected subsets with the overall data pool, focusing solely on\ndiversity. In this paper, we propose a Bi-Level Active Finetuning framework to\nselect the samples for annotation in one shot, which includes two stages: core\nsample selection for diversity, and boundary sample selection for uncertainty.\nThe process begins with the identification of pseudo-class centers, followed by\nan innovative denoising method and an iterative strategy for boundary sample\nselection in the high-dimensional feature space, all without relying on\nground-truth labels. Our comprehensive experiments provide both qualitative and\nquantitative evidence of our method's efficacy, outperforming all the existing\nbaselines.\n", "date": "2024-03-15", "categories": []}
{"id": "2211.11612", "title": "Plug and Play Active Learning for Object Detection", "abstract": "  Annotating datasets for object detection is an expensive and time-consuming\nendeavor. To minimize this burden, active learning (AL) techniques are employed\nto select the most informative samples for annotation within a constrained\n\"annotation budget\". Traditional AL strategies typically rely on model\nuncertainty or sample diversity for query sampling, while more advanced methods\nhave focused on developing AL-specific object detector architectures to enhance\nperformance. However, these specialized approaches are not readily adaptable to\ndifferent object detectors due to the significant engineering effort required\nfor integration. To overcome this challenge, we introduce Plug and Play Active\nLearning (PPAL), a simple and effective AL strategy for object detection. PPAL\nis a two-stage method comprising uncertainty-based and diversity-based sampling\nphases. In the first stage, our Difficulty Calibrated Uncertainty Sampling\nleverage a category-wise difficulty coefficient that combines both\nclassification and localisation difficulties to re-weight instance\nuncertainties, from which we sample a candidate pool for the subsequent\ndiversity-based sampling. In the second stage, we propose Category Conditioned\nMatching Similarity to better compute the similarities of multi-instance images\nas ensembles of their instance similarities, which is used by the k-Means++\nalgorithm to sample the final AL queries. PPAL makes no change to model\narchitectures or detector training pipelines; hence it can be easily\ngeneralized to different object detectors. We benchmark PPAL on the MS-COCO and\nPascal VOC datasets using different detector architectures and show that our\nmethod outperforms prior work by a large margin. Code is available at\nhttps://github.com/ChenhongyiYang/PPAL\n", "date": "2022-11-21", "categories": []}
{"id": "2310.08035", "title": "BaSAL: Size-Balanced Warm Start Active Learning for LiDAR Semantic\n  Segmentation", "abstract": "  Active learning strives to reduce the need for costly data annotation, by\nrepeatedly querying an annotator to label the most informative samples from a\npool of unlabeled data, and then training a model from these samples. We\nidentify two problems with existing active learning methods for LiDAR semantic\nsegmentation. First, they overlook the severe class imbalance inherent in LiDAR\nsemantic segmentation datasets. Second, to bootstrap the active learning loop\nwhen there is no labeled data available, they train their initial model from\nrandomly selected data samples, leading to low performance. This situation is\nreferred to as the cold start problem. To address these problems we propose\nBaSAL, a size-balanced warm start active learning model, based on the\nobservation that each object class has a characteristic size. By sampling\nobject clusters according to their size, we can thus create a size-balanced\ndataset that is also more class-balanced. Furthermore, in contrast to existing\ninformation measures like entropy or CoreSet, size-based sampling does not\nrequire a pretrained model, thus addressing the cold start problem effectively.\nResults show that we are able to improve the performance of the initial model\nby a large margin. Combining warm start and size-balanced sampling with\nestablished information measures, our approach achieves comparable performance\nto training on the entire SemanticKITTI dataset, despite using only 5% of the\nannotations, outperforming existing active learning methods. We also match the\nexisting state-of-the-art in active learning on nuScenes. Our code is available\nat: https://github.com/Tony-WJR/BaSAL.\n", "date": "2023-10-12", "categories": []}
{"id": "2310.14230", "title": "A comprehensive survey on deep active learning in medical image analysis", "abstract": "  Deep learning has achieved widespread success in medical image analysis,\nleading to an increasing demand for large-scale expert-annotated medical image\ndatasets. Yet, the high cost of annotating medical images severely hampers the\ndevelopment of deep learning in this field. To reduce annotation costs, active\nlearning aims to select the most informative samples for annotation and train\nhigh-performance models with as few labeled samples as possible. In this\nsurvey, we review the core methods of active learning, including the evaluation\nof informativeness and sampling strategy. For the first time, we provide a\ndetailed summary of the integration of active learning with other\nlabel-efficient techniques, such as semi-supervised, self-supervised learning,\nand so on. We also summarize active learning works that are specifically\ntailored to medical image analysis. Additionally, we conduct a thorough\ncomparative analysis of the performance of different AL methods in medical\nimage analysis with experiments. In the end, we offer our perspectives on the\nfuture trends and challenges of active learning and its applications in medical\nimage analysis.\n", "date": "2023-10-22", "categories": []}
{"id": "2312.08150", "title": "Active learning with biased non-response to label requests", "abstract": "  Active learning can improve the efficiency of training prediction models by\nidentifying the most informative new labels to acquire. However, non-response\nto label requests can impact active learning's effectiveness in real-world\ncontexts. We conceptualise this degradation by considering the type of\nnon-response present in the data, demonstrating that biased non-response is\nparticularly detrimental to model performance. We argue that biased\nnon-response is likely in contexts where the labelling process, by nature,\nrelies on user interactions. To mitigate the impact of biased non-response, we\npropose a cost-based correction to the sampling strategy--the Upper Confidence\nBound of the Expected Utility (UCB-EU)--that can, plausibly, be applied to any\nactive learning algorithm. Through experiments, we demonstrate that our method\nsuccessfully reduces the harm from labelling non-response in many settings.\nHowever, we also characterise settings where the non-response bias in the\nannotations remains detrimental under UCB-EU for specific sampling methods and\ndata generating processes. Finally, we evaluate our method on a real-world\ndataset from an e-commerce platform. We show that UCB-EU yields substantial\nperformance improvements to conversion models that are trained on clicked\nimpressions. Most generally, this research serves to both better conceptualise\nthe interplay between types of non-response and model improvements via active\nlearning, and to provide a practical, easy-to-implement correction that\nmitigates model degradation.\n", "date": "2023-12-13", "categories": []}
{"id": "2401.04305", "title": "Advancing Deep Active Learning & Data Subset Selection: Unifying\n  Principles with Information-Theory Intuitions", "abstract": "  At its core, this thesis aims to enhance the practicality of deep learning by\nimproving the label and training efficiency of deep learning models. To this\nend, we investigate data subset selection techniques, specifically active\nlearning and active sampling, grounded in information-theoretic principles.\nActive learning improves label efficiency, while active sampling enhances\ntraining efficiency. Supervised deep learning models often require extensive\ntraining with labeled data. Label acquisition can be expensive and\ntime-consuming, and training large models is resource-intensive, hindering the\nadoption outside academic research and \"big tech.\" Existing methods for data\nsubset selection in deep learning often rely on heuristics or lack a principled\ninformation-theoretic foundation. In contrast, this thesis examines several\nobjectives for data subset selection and their applications within deep\nlearning, striving for a more principled approach inspired by information\ntheory. We begin by disentangling epistemic and aleatoric uncertainty in single\nforward-pass deep neural networks, which provides helpful intuitions and\ninsights into different forms of uncertainty and their relevance for data\nsubset selection. We then propose and investigate various approaches for active\nlearning and data subset selection in (Bayesian) deep learning. Finally, we\nrelate various existing and proposed approaches to approximations of\ninformation quantities in weight or prediction space. Underpinning this work is\na principled and practical notation for information-theoretic quantities that\nincludes both random variables and observed outcomes. This thesis demonstrates\nthe benefits of working from a unified perspective and highlights the potential\nimpact of our contributions to the practical application of deep learning.\n", "date": "2024-01-09", "categories": []}
{"id": "2402.18064", "title": "Automated Testing of Spatially-Dependent Environmental Hypotheses\n  through Active Transfer Learning", "abstract": "  The efficient collection of samples is an important factor in outdoor\ninformation gathering applications on account of high sampling costs such as\ntime, energy, and potential destruction to the environment. Utilization of\navailable a-priori data can be a powerful tool for increasing efficiency.\nHowever, the relationships of this data with the quantity of interest are often\nnot known ahead of time, limiting the ability to leverage this knowledge for\nimproved planning efficiency. To this end, this work combines transfer learning\nand active learning through a Multi-Task Gaussian Process and an\ninformation-based objective function. Through this combination it can explore\nthe space of hypothetical inter-quantity relationships and evaluate these\nhypotheses in real-time, allowing this new knowledge to be immediately\nexploited for future plans. The performance of the proposed method is evaluated\nagainst synthetic data and is shown to evaluate multiple hypotheses correctly.\nIts effectiveness is also demonstrated on real datasets. The technique is able\nto identify and leverage hypotheses which show a medium or strong correlation\nto reduce prediction error by a factor of 1.4--3.4 within the first 7 samples,\nand poor hypotheses are quickly identified and rejected eventually having no\nadverse effect.\n", "date": "2024-02-28", "categories": []}
{"id": "2403.03741", "title": "SUPClust: Active Learning at the Boundaries", "abstract": "  Active learning is a machine learning paradigm designed to optimize model\nperformance in a setting where labeled data is expensive to acquire. In this\nwork, we propose a novel active learning method called SUPClust that seeks to\nidentify points at the decision boundary between classes. By targeting these\npoints, SUPClust aims to gather information that is most informative for\nrefining the model's prediction of complex decision regions. We demonstrate\nexperimentally that labeling these points leads to strong model performance.\nThis improvement is observed even in scenarios characterized by strong class\nimbalance.\n", "date": "2024-03-06", "categories": []}
{"id": "2403.02877", "title": "ActiveAD: Planning-Oriented Active Learning for End-to-End Autonomous\n  Driving", "abstract": "  End-to-end differentiable learning for autonomous driving (AD) has recently\nbecome a prominent paradigm. One main bottleneck lies in its voracious appetite\nfor high-quality labeled data e.g. 3D bounding boxes and semantic segmentation,\nwhich are notoriously expensive to manually annotate. The difficulty is further\npronounced due to the prominent fact that the behaviors within samples in AD\noften suffer from long tailed distribution. In other words, a large part of\ncollected data can be trivial (e.g. simply driving forward in a straight road)\nand only a few cases are safety-critical. In this paper, we explore a\npractically important yet under-explored problem about how to achieve sample\nand label efficiency for end-to-end AD. Specifically, we design a\nplanning-oriented active learning method which progressively annotates part of\ncollected raw data according to the proposed diversity and usefulness criteria\nfor planning routes. Empirically, we show that our planning-oriented approach\ncould outperform general active learning methods by a large margin. Notably,\nour method achieves comparable performance with state-of-the-art end-to-end AD\nmethods - by using only 30% nuScenes data. We hope our work could inspire\nfuture works to explore end-to-end AD from a data-centric perspective in\naddition to methodology efforts.\n", "date": "2024-03-05", "categories": []}
{"id": "2306.09910", "title": "LabelBench: A Comprehensive Framework for Benchmarking Adaptive\n  Label-Efficient Learning", "abstract": "  Labeled data are critical to modern machine learning applications, but\nobtaining labels can be expensive. To mitigate this cost, machine learning\nmethods, such as transfer learning, semi-supervised learning and active\nlearning, aim to be label-efficient: achieving high predictive performance from\nrelatively few labeled examples. While obtaining the best label-efficiency in\npractice often requires combinations of these techniques, existing benchmark\nand evaluation frameworks do not capture a concerted combination of all such\ntechniques. This paper addresses this deficiency by introducing LabelBench, a\nnew computationally-efficient framework for joint evaluation of multiple\nlabel-efficient learning techniques. As an application of LabelBench, we\nintroduce a novel benchmark of state-of-the-art active learning methods in\ncombination with semi-supervised learning for fine-tuning pretrained vision\ntransformers. Our benchmark demonstrates better label-efficiencies than\npreviously reported in active learning. LabelBench's modular codebase is\nopen-sourced for the broader community to contribute label-efficient learning\nmethods and benchmarks. The repository can be found at:\nhttps://github.com/EfficientTraining/LabelBench.\n", "date": "2023-06-16", "categories": []}
{"id": "2403.01346", "title": "Improve Cost Efficiency of Active Learning over Noisy Dataset", "abstract": "  Active learning is a learning strategy whereby the machine learning algorithm\nactively identifies and labels data points to optimize its learning. This\nstrategy is particularly effective in domains where an abundance of unlabeled\ndata exists, but the cost of labeling these data points is prohibitively\nexpensive. In this paper, we consider cases of binary classification, where\nacquiring a positive instance incurs a significantly higher cost compared to\nthat of negative instances. For example, in the financial industry, such as in\nmoney-lending businesses, a defaulted loan constitutes a positive event leading\nto substantial financial loss. To address this issue, we propose a shifted\nnormal distribution sampling function that samples from a wider range than\ntypical uncertainty sampling. Our simulation underscores that our proposed\nsampling function limits both noisy and positive label selection, delivering\nbetween 20% and 32% improved cost efficiency over different test datasets.\n", "date": "2024-03-02", "categories": []}
{"id": "2403.01352", "title": "Improving Uncertainty Sampling with Bell Curve Weight Function", "abstract": "  Typically, a supervised learning model is trained using passive learning by\nrandomly selecting unlabelled instances to annotate. This approach is effective\nfor learning a model, but can be costly in cases where acquiring labelled\ninstances is expensive. For example, it can be time-consuming to manually\nidentify spam mails (labelled instances) from thousands of emails (unlabelled\ninstances) flooding an inbox during initial data collection. Generally, we\nanswer the above scenario with uncertainty sampling, an active learning method\nthat improves the efficiency of supervised learning by using fewer labelled\ninstances than passive learning. Given an unlabelled data pool, uncertainty\nsampling queries the labels of instances where the predicted probabilities, p,\nfall into the uncertainty region, i.e., $p \\approx 0.5$. The newly acquired\nlabels are then added to the existing labelled data pool to learn a new model.\nNonetheless, the performance of uncertainty sampling is susceptible to the area\nof unpredictable responses (AUR) and the nature of the dataset. It is difficult\nto determine whether to use passive learning or uncertainty sampling without\nprior knowledge of a new dataset. To address this issue, we propose bell curve\nsampling, which employs a bell curve weight function to acquire new labels.\nWith the bell curve centred at p=0.5, bell curve sampling selects instances\nwhose predicted values are in the uncertainty area most of the time without\nneglecting the rest. Simulation results show that, most of the time bell curve\nsampling outperforms uncertainty sampling and passive learning in datasets of\ndifferent natures and with AUR.\n", "date": "2024-03-03", "categories": []}
{"id": "2304.03870", "title": "ASPEST: Bridging the Gap Between Active Learning and Selective\n  Prediction", "abstract": "  Selective prediction aims to learn a reliable model that abstains from making\npredictions when uncertain. These predictions can then be deferred to humans\nfor further evaluation. As an everlasting challenge for machine learning, in\nmany real-world scenarios, the distribution of test data is different from the\ntraining data. This results in more inaccurate predictions, and often increased\ndependence on humans, which can be difficult and expensive. Active learning\naims to lower the overall labeling effort, and hence human dependence, by\nquerying the most informative examples. Selective prediction and active\nlearning have been approached from different angles, with the connection\nbetween them missing. In this work, we introduce a new learning paradigm,\nactive selective prediction, which aims to query more informative samples from\nthe shifted target domain while increasing accuracy and coverage. For this new\nparadigm, we propose a simple yet effective approach, ASPEST, that utilizes\nensembles of model snapshots with self-training with their aggregated outputs\nas pseudo labels. Extensive experiments on numerous image, text and structured\ndatasets, which suffer from domain shifts, demonstrate that ASPEST can\nsignificantly outperform prior work on selective prediction and active learning\n(e.g. on the MNIST$\\to$SVHN benchmark with the labeling budget of 100, ASPEST\nimproves the AUACC metric from 79.36% to 88.84%) and achieves more optimal\nutilization of humans in the loop.\n", "date": "2023-04-07", "categories": []}
{"id": "2401.15721", "title": "A Study of Acquisition Functions for Medical Imaging Deep Active\n  Learning", "abstract": "  The Deep Learning revolution has enabled groundbreaking achievements in\nrecent years. From breast cancer detection to protein folding, deep learning\nalgorithms have been at the core of very important advancements. However, these\nmodern advancements are becoming more and more data-hungry, especially on\nlabeled data whose availability is scarce: this is even more prevalent in the\nmedical context. In this work, we show how active learning could be very\neffective in data scarcity situations, where obtaining labeled data (or\nannotation budget is very limited). We compare several selection criteria\n(BALD, MeanSTD, and MaxEntropy) on the ISIC 2016 dataset. We also explored the\neffect of acquired pool size on the model's performance. Our results suggest\nthat uncertainty is useful to the Melanoma detection task, and confirms the\nhypotheses of the author of the paper of interest, that \\textit{bald} performs\non average better than other acquisition functions. Our extended analyses\nhowever revealed that all acquisition functions perform badly on the positive\n(cancerous) samples, suggesting exploitation of class unbalance, which could be\ncrucial in real-world settings. We finish by suggesting future work directions\nthat would be useful to improve this current work. The code of our\nimplementation is open-sourced at\n\\url{https://github.com/bonaventuredossou/ece526_course_project}\n", "date": "2024-01-28", "categories": []}
{"id": "2311.07634", "title": "ActiveDC: Distribution Calibration for Active Finetuning", "abstract": "  The pretraining-finetuning paradigm has gained popularity in various computer\nvision tasks. In this paradigm, the emergence of active finetuning arises due\nto the abundance of large-scale data and costly annotation requirements. Active\nfinetuning involves selecting a subset of data from an unlabeled pool for\nannotation, facilitating subsequent finetuning. However, the use of a limited\nnumber of training samples can lead to a biased distribution, potentially\nresulting in model overfitting. In this paper, we propose a new method called\nActiveDC for the active finetuning tasks. Firstly, we select samples for\nannotation by optimizing the distribution similarity between the subset to be\nselected and the entire unlabeled pool in continuous space. Secondly, we\ncalibrate the distribution of the selected samples by exploiting implicit\ncategory information in the unlabeled pool. The feature visualization provides\nan intuitive sense of the effectiveness of our approach to distribution\ncalibration. We conducted extensive experiments on three image classification\ndatasets with different sampling ratios. The results indicate that ActiveDC\nconsistently outperforms the baseline performance in all image classification\ntasks. The improvement is particularly significant when the sampling ratio is\nlow, with performance gains of up to 10%. Our code will be released.\n", "date": "2023-11-13", "categories": []}
{"id": "2402.15613", "title": "Towards Efficient Active Learning in NLP via Pretrained Representations", "abstract": "  Fine-tuning Large Language Models (LLMs) is now a common approach for text\nclassification in a wide range of applications. When labeled documents are\nscarce, active learning helps save annotation efforts but requires retraining\nof massive models on each acquisition iteration. We drastically expedite this\nprocess by using pretrained representations of LLMs within the active learning\nloop and, once the desired amount of labeled data is acquired, fine-tuning that\nor even a different pretrained LLM on this labeled data to achieve the best\nperformance. As verified on common text classification benchmarks with\npretrained BERT and RoBERTa as the backbone, our strategy yields similar\nperformance to fine-tuning all the way through the active learning loop but is\norders of magnitude less computationally expensive. The data acquired with our\nprocedure generalizes across pretrained networks, allowing flexibility in\nchoosing the final model or updating it as newer versions get released.\n", "date": "2024-02-23", "categories": []}
{"id": "2402.15757", "title": "Batch Active Learning of Reward Functions from Human Preferences", "abstract": "  Data generation and labeling are often expensive in robot learning.\nPreference-based learning is a concept that enables reliable labeling by\nquerying users with preference questions. Active querying methods are commonly\nemployed in preference-based learning to generate more informative data at the\nexpense of parallelization and computation time. In this paper, we develop a\nset of novel algorithms, batch active preference-based learning methods, that\nenable efficient learning of reward functions using as few data samples as\npossible while still having short query generation times and also retaining\nparallelizability. We introduce a method based on determinantal point processes\n(DPP) for active batch generation and several heuristic-based alternatives.\nFinally, we present our experimental results for a variety of robotics tasks in\nsimulation. Our results suggest that our batch active learning algorithm\nrequires only a few queries that are computed in a short amount of time. We\nshowcase one of our algorithms in a study to learn human users' preferences.\n", "date": "2024-02-24", "categories": []}
{"id": "2402.16237", "title": "Active Level Set Estimation for Continuous Search Space with Theoretical\n  Guarantee", "abstract": "  A common problem encountered in many real-world applications is level set\nestimation where the goal is to determine the region in the function domain\nwhere the function is above or below a given threshold. When the function is\nblack-box and expensive to evaluate, the level sets need to be found in a\nminimum set of function evaluations. Existing methods often assume a discrete\nsearch space with a finite set of data points for function evaluations and\nestimating the level sets. When applied to a continuous search space, these\nmethods often need to first discretize the space which leads to poor results\nwhile needing high computational time. While some methods cater for the\ncontinuous setting, they still lack a proper guarantee for theoretical\nconvergence. To address this problem, we propose a novel algorithm that does\nnot need any discretization and can directly work in continuous search spaces.\nOur method suggests points by constructing an acquisition function that is\ndefined as a measure of confidence of the function being higher or lower than\nthe given threshold. A theoretical analysis for the convergence of the\nalgorithm to an accurate solution is provided. On multiple synthetic and\nreal-world datasets, our algorithm successfully outperforms state-of-the-art\nmethods.\n", "date": "2024-02-26", "categories": []}
{"id": "2211.12620", "title": "Promises and Pitfalls of Threshold-based Auto-labeling", "abstract": "  Creating large-scale high-quality labeled datasets is a major bottleneck in\nsupervised machine learning workflows. Threshold-based auto-labeling (TBAL),\nwhere validation data obtained from humans is used to find a confidence\nthreshold above which the data is machine-labeled, reduces reliance on manual\nannotation. TBAL is emerging as a widely-used solution in practice. Given the\nlong shelf-life and diverse usage of the resulting datasets, understanding when\nthe data obtained by such auto-labeling systems can be relied on is crucial.\nThis is the first work to analyze TBAL systems and derive sample complexity\nbounds on the amount of human-labeled validation data required for guaranteeing\nthe quality of machine-labeled data. Our results provide two crucial insights.\nFirst, reasonable chunks of unlabeled data can be automatically and accurately\nlabeled by seemingly bad models. Second, a hidden downside of TBAL systems is\npotentially prohibitive validation data usage. Together, these insights\ndescribe the promise and pitfalls of using such systems. We validate our\ntheoretical guarantees with extensive experiments on synthetic and real\ndatasets.\n", "date": "2022-11-22", "categories": []}
{"id": "2402.11223", "title": "HEAL: Brain-inspired Hyperdimensional Efficient Active Learning", "abstract": "  Drawing inspiration from the outstanding learning capability of our human\nbrains, Hyperdimensional Computing (HDC) emerges as a novel computing paradigm,\nand it leverages high-dimensional vector presentation and operations for\nbrain-like lightweight Machine Learning (ML). Practical deployments of HDC have\nsignificantly enhanced the learning efficiency compared to current deep ML\nmethods on a broad spectrum of applications. However, boosting the data\nefficiency of HDC classifiers in supervised learning remains an open question.\nIn this paper, we introduce Hyperdimensional Efficient Active Learning (HEAL),\na novel Active Learning (AL) framework tailored for HDC classification. HEAL\nproactively annotates unlabeled data points via uncertainty and\ndiversity-guided acquisition, leading to a more efficient dataset annotation\nand lowering labor costs. Unlike conventional AL methods that only support\nclassifiers built upon deep neural networks (DNN), HEAL operates without the\nneed for gradient or probabilistic computations. This allows it to be\neffortlessly integrated with any existing HDC classifier architecture. The key\ndesign of HEAL is a novel approach for uncertainty estimation in HDC\nclassifiers through a lightweight HDC ensemble with prior hypervectors.\nAdditionally, by exploiting hypervectors as prototypes (i.e., compact\nrepresentations), we develop an extra metric for HEAL to select diverse samples\nwithin each batch for annotation. Our evaluation shows that HEAL surpasses a\ndiverse set of baselines in AL quality and achieves notably faster acquisition\nthan many BNN-powered or diversity-guided AL methods, recording 11 times to\n40,000 times speedup in acquisition runtime per batch.\n", "date": "2024-02-17", "categories": []}
{"id": "2402.11973", "title": "Bayesian Active Learning for Censored Regression", "abstract": "  Bayesian active learning is based on information theoretical approaches that\nfocus on maximising the information that new observations provide to the model\nparameters. This is commonly done by maximising the Bayesian Active Learning by\nDisagreement (BALD) acquisitions function. However, we highlight that it is\nchallenging to estimate BALD when the new data points are subject to\ncensorship, where only clipped values of the targets are observed. To address\nthis, we derive the entropy and the mutual information for censored\ndistributions and derive the BALD objective for active learning in censored\nregression ($\\mathcal{C}$-BALD). We propose a novel modelling approach to\nestimate the $\\mathcal{C}$-BALD objective and use it for active learning in the\ncensored setting. Across a wide range of datasets and models, we demonstrate\nthat $\\mathcal{C}$-BALD outperforms other Bayesian active learning methods in\ncensored regression.\n", "date": "2024-02-19", "categories": []}
{"id": "2304.11005", "title": "Self-Correcting Bayesian Optimization through Bayesian Active Learning", "abstract": "  Gaussian processes are the model of choice in Bayesian optimization and\nactive learning. Yet, they are highly dependent on cleverly chosen\nhyperparameters to reach their full potential, and little effort is devoted to\nfinding good hyperparameters in the literature. We demonstrate the impact of\nselecting good hyperparameters for GPs and present two acquisition functions\nthat explicitly prioritize hyperparameter learning. Statistical distance-based\nActive Learning (SAL) considers the average disagreement between samples from\nthe posterior, as measured by a statistical distance. SAL outperforms the\nstate-of-the-art in Bayesian active learning on several test functions. We then\nintroduce Self-Correcting Bayesian Optimization (SCoreBO), which extends SAL to\nperform Bayesian optimization and active learning simultaneously. SCoreBO\nlearns the model hyperparameters at improved rates compared to vanilla BO,\nwhile outperforming the latest Bayesian optimization methods on traditional\nbenchmarks. Moreover, we demonstrate the importance of self-correction on\natypical Bayesian optimization tasks.\n", "date": "2023-04-21", "categories": []}
{"id": "2302.10295", "title": "Correlation Clustering with Active Learning of Pairwise Similarities", "abstract": "  Correlation clustering is a well-known unsupervised learning setting that\ndeals with positive and negative pairwise similarities. In this paper, we study\nthe case where the pairwise similarities are not given in advance and must be\nqueried in a cost-efficient way. Thereby, we develop a generic active learning\nframework for this task that benefits from several advantages, e.g.,\nflexibility in the type of feedback that a user/annotator can provide,\nadaptation to any correlation clustering algorithm and query strategy, and\nrobustness to noise. In addition, we propose and analyze a number of novel\nquery strategies suited to this setting. We demonstrate the effectiveness of\nour framework and the proposed query strategies via several experimental\nstudies.\n", "date": "2023-02-20", "categories": []}
{"id": "2402.02110", "title": "Composite Active Learning: Towards Multi-Domain Active Learning with\n  Theoretical Guarantees", "abstract": "  Active learning (AL) aims to improve model performance within a fixed\nlabeling budget by choosing the most informative data points to label. Existing\nAL focuses on the single-domain setting, where all data come from the same\ndomain (e.g., the same dataset). However, many real-world tasks often involve\nmultiple domains. For example, in visual recognition, it is often desirable to\ntrain an image classifier that works across different environments (e.g.,\ndifferent backgrounds), where images from each environment constitute one\ndomain. Such a multi-domain AL setting is challenging for prior methods because\nthey (1) ignore the similarity among different domains when assigning labeling\nbudget and (2) fail to handle distribution shift of data across different\ndomains. In this paper, we propose the first general method, dubbed composite\nactive learning (CAL), for multi-domain AL. Our approach explicitly considers\nthe domain-level and instance-level information in the problem; CAL first\nassigns domain-level budgets according to domain-level importance, which is\nestimated by optimizing an upper error bound that we develop; with the\ndomain-level budgets, CAL then leverages a certain instance-level query\nstrategy to select samples to label from each domain. Our theoretical analysis\nshows that our method achieves a better error bound compared to current AL\nmethods. Our empirical results demonstrate that our approach significantly\noutperforms the state-of-the-art AL methods on both synthetic and real-world\nmulti-domain datasets. Code is available at\nhttps://github.com/Wang-ML-Lab/multi-domain-active-learning.\n", "date": "2024-02-03", "categories": []}
{"id": "2402.06045", "title": "Direct Acquisition Optimization for Low-Budget Active Learning", "abstract": "  Active Learning (AL) has gained prominence in integrating data-intensive\nmachine learning (ML) models into domains with limited labeled data. However,\nits effectiveness diminishes significantly when the labeling budget is low. In\nthis paper, we first empirically observe the performance degradation of\nexisting AL algorithms in the low-budget settings, and then introduce Direct\nAcquisition Optimization (DAO), a novel AL algorithm that optimizes sample\nselections based on expected true loss reduction. Specifically, DAO utilizes\ninfluence functions to update model parameters and incorporates an additional\nacquisition strategy to mitigate bias in loss estimation. This approach\nfacilitates a more accurate estimation of the overall error reduction, without\nextensive computations or reliance on labeled data. Experiments demonstrate\nDAO's effectiveness in low budget settings, outperforming state-of-the-arts\napproaches across seven benchmarks.\n", "date": "2024-02-08", "categories": []}
{"id": "2402.06056", "title": "ActiveDP: Bridging Active Learning and Data Programming", "abstract": "  Modern machine learning models require large labelled datasets to achieve\ngood performance, but manually labelling large datasets is expensive and\ntime-consuming. The data programming paradigm enables users to label large\ndatasets efficiently but produces noisy labels, which deteriorates the\ndownstream model's performance. The active learning paradigm, on the other\nhand, can acquire accurate labels but only for a small fraction of instances.\nIn this paper, we propose ActiveDP, an interactive framework bridging active\nlearning and data programming together to generate labels with both high\naccuracy and coverage, combining the strengths of both paradigms. Experiments\nshow that ActiveDP outperforms previous weak supervision and active learning\napproaches and consistently performs well under different labelling budgets.\n", "date": "2024-02-08", "categories": []}
{"id": "2402.06276", "title": "Safe Active Learning for Time-Series Modeling with Gaussian Processes", "abstract": "  Learning time-series models is useful for many applications, such as\nsimulation and forecasting. In this study, we consider the problem of actively\nlearning time-series models while taking given safety constraints into account.\nFor time-series modeling we employ a Gaussian process with a nonlinear\nexogenous input structure. The proposed approach generates data appropriate for\ntime series model learning, i.e. input and output trajectories, by dynamically\nexploring the input space. The approach parametrizes the input trajectory as\nconsecutive trajectory sections, which are determined stepwise given safety\nrequirements and past observations. We analyze the proposed algorithm and\nevaluate it empirically on a technical application. The results show the\neffectiveness of our approach in a realistic technical use case.\n", "date": "2024-02-09", "categories": []}
{"id": "2402.06560", "title": "Video Annotator: A framework for efficiently building video classifiers\n  using vision-language models and active learning", "abstract": "  High-quality and consistent annotations are fundamental to the successful\ndevelopment of robust machine learning models. Traditional data annotation\nmethods are resource-intensive and inefficient, often leading to a reliance on\nthird-party annotators who are not the domain experts. Hard samples, which are\nusually the most informative for model training, tend to be difficult to label\naccurately and consistently without business context. These can arise\nunpredictably during the annotation process, requiring a variable number of\niterations and rounds of feedback, leading to unforeseen expenses and time\ncommitments to guarantee quality.\n  We posit that more direct involvement of domain experts, using a\nhuman-in-the-loop system, can resolve many of these practical challenges. We\npropose a novel framework we call Video Annotator (VA) for annotating,\nmanaging, and iterating on video classification datasets. Our approach offers a\nnew paradigm for an end-user-centered model development process, enhancing the\nefficiency, usability, and effectiveness of video classifiers. Uniquely, VA\nallows for a continuous annotation process, seamlessly integrating data\ncollection and model training.\n  We leverage the zero-shot capabilities of vision-language foundation models\ncombined with active learning techniques, and demonstrate that VA enables the\nefficient creation of high-quality models. VA achieves a median 6.8 point\nimprovement in Average Precision relative to the most competitive baseline\nacross a wide-ranging assortment of tasks. We release a dataset with 153k\nlabels across 56 video understanding tasks annotated by three professional\nvideo editors using VA, and also release code to replicate our experiments at:\nhttp://github.com/netflix/videoannotator.\n", "date": "2024-02-09", "categories": []}
{"id": "2309.05853", "title": "ChemSpaceAL: An Efficient Active Learning Methodology Applied to\n  Protein-Specific Molecular Generation", "abstract": "  The incredible capabilities of generative artificial intelligence models have\ninevitably led to their application in the domain of drug discovery. Within\nthis domain, the vastness of chemical space motivates the development of more\nefficient methods for identifying regions with molecules that exhibit desired\ncharacteristics. In this work, we present a computationally efficient active\nlearning methodology that requires evaluation of only a subset of the generated\ndata in the constructed sample space to successfully align a generative model\nwith respect to a specified objective. We demonstrate the applicability of this\nmethodology to targeted molecular generation by fine-tuning a GPT-based\nmolecular generator toward a protein with FDA-approved small-molecule\ninhibitors, c-Abl kinase. Remarkably, the model learns to generate molecules\nsimilar to the inhibitors without prior knowledge of their existence, and even\nreproduces two of them exactly. We also show that the methodology is effective\nfor a protein without any commercially available small-molecule inhibitors, the\nHNH domain of the CRISPR-associated protein 9 (Cas9) enzyme. We believe that\nthe inherent generality of this method ensures that it will remain applicable\nas the exciting field of in silico molecular generation evolves. To facilitate\nimplementation and reproducibility, we have made all of our software available\nthrough the open-source ChemSpaceAL Python package.\n", "date": "2023-09-11", "categories": []}
{"id": "2402.04896", "title": "Learning from the Best: Active Learning for Wireless Communications", "abstract": "  Collecting an over-the-air wireless communications training dataset for deep\nlearning-based communication tasks is relatively simple. However, labeling the\ndataset requires expert involvement and domain knowledge, may involve private\nintellectual properties, and is often computationally and financially\nexpensive. Active learning is an emerging area of research in machine learning\nthat aims to reduce the labeling overhead without accuracy degradation. Active\nlearning algorithms identify the most critical and informative samples in an\nunlabeled dataset and label only those samples, instead of the complete set. In\nthis paper, we introduce active learning for deep learning applications in\nwireless communications, and present its different categories. We present a\ncase study of deep learning-based mmWave beam selection, where labeling is\nperformed by a compute-intensive algorithm based on exhaustive search. We\nevaluate the performance of different active learning algorithms on a publicly\navailable multi-modal dataset with different modalities including image and\nLiDAR. Our results show that using an active learning algorithm for\nclass-imbalanced datasets can reduce labeling overhead by up to 50% for this\ndataset while maintaining the same accuracy as classical training.\n", "date": "2024-01-23", "categories": []}
{"id": "2402.03252", "title": "Fair Active Ranking from Pairwise Preferences", "abstract": "  We investigate the problem of probably approximately correct and fair (PACF)\nranking of items by adaptively evoking pairwise comparisons. Given a set of $n$\nitems that belong to disjoint groups, our goal is to find an $(\\epsilon,\n\\delta)$-PACF-Ranking according to a fair objective function that we propose.\nWe assume access to an oracle, wherein, for each query, the learner can choose\na pair of items and receive stochastic winner feedback from the oracle. Our\nproposed objective function asks to minimize the $\\ell_q$ norm of the error of\nthe groups, where the error of a group is the $\\ell_p$ norm of the error of all\nthe items within that group, for $p, q \\geq 1$. This generalizes the objective\nfunction of $\\epsilon$-Best-Ranking, proposed by Saha & Gopalan (2019).\n  By adopting our objective function, we gain the flexibility to explore\nfundamental fairness concepts like equal or proportionate errors within a\nunified framework. Adjusting parameters $p$ and $q$ allows tailoring to\nspecific fairness preferences. We present both group-blind and group-aware\nalgorithms and analyze their sample complexity. We provide matching lower\nbounds up to certain logarithmic factors for group-blind algorithms. For a\nrestricted class of group-aware algorithms, we show that we can get reasonable\nlower bounds. We conduct comprehensive experiments on both real-world and\nsynthetic datasets to complement our theoretical findings.\n", "date": "2024-02-05", "categories": []}
{"id": "2402.02321", "title": "Active Learning for Graphs with Noisy Structures", "abstract": "  Graph Neural Networks (GNNs) have seen significant success in tasks such as\nnode classification, largely contingent upon the availability of sufficient\nlabeled nodes. Yet, the excessive cost of labeling large-scale graphs led to a\nfocus on active learning on graphs, which aims for effective data selection to\nmaximize downstream model performance. Notably, most existing methods assume\nreliable graph topology, while real-world scenarios often present noisy graphs.\nGiven this, designing a successful active learning framework for noisy graphs\nis highly needed but challenging, as selecting data for labeling and obtaining\na clean graph are two tasks naturally interdependent: selecting high-quality\ndata requires clean graph structure while cleaning noisy graph structure\nrequires sufficient labeled data. Considering the complexity mentioned above,\nwe propose an active learning framework, GALClean, which has been specifically\ndesigned to adopt an iterative approach for conducting both data selection and\ngraph purification simultaneously with best information learned from the prior\niteration. Importantly, we summarize GALClean as an instance of the\nExpectation-Maximization algorithm, which provides a theoretical understanding\nof its design and mechanisms. This theory naturally leads to an enhanced\nversion, GALClean+. Extensive experiments have demonstrated the effectiveness\nand robustness of our proposed method across various types and levels of noisy\ngraphs.\n", "date": "2024-02-04", "categories": []}
{"id": "2402.02516", "title": "Adaptive scheduling for adaptive sampling in POS taggers construction", "abstract": "  We introduce an adaptive scheduling for adaptive sampling as a novel way of\nmachine learning in the construction of part-of-speech taggers. The goal is to\nspeed up the training on large data sets, without significant loss of\nperformance with regard to an optimal configuration. In contrast to previous\nmethods using a random, fixed or regularly rising spacing between the\ninstances, ours analyzes the shape of the learning curve geometrically in\nconjunction with a functional model to increase or decrease it at any time. The\nalgorithm proves to be formally correct regarding our working hypotheses.\nNamely, given a case, the following one is the nearest ensuring a net gain of\nlearning ability from the former, it being possible to modulate the level of\nrequirement for this condition. We also improve the robustness of sampling by\npaying greater attention to those regions of the training data base subject to\na temporary inflation in performance, thus preventing the learning from\nstopping prematurely.\n  The proposal has been evaluated on the basis of its reliability to identify\nthe convergence of models, corroborating our expectations. While a concrete\nhalting condition is used for testing, users can choose any condition\nwhatsoever to suit their own specific needs.\n", "date": "2024-02-04", "categories": []}
{"id": "2402.01577", "title": "Deep Active Learning for Data Mining from Conflict Text Corpora", "abstract": "  High-resolution event data on armed conflict and related processes have\nrevolutionized the study of political contention with datasets like UCDP GED,\nACLED etc. However, most of these datasets limit themselves to collecting\nspatio-temporal (high-resolution) and intensity data. Information on dynamics,\nsuch as targets, tactics, purposes etc. are rarely collected owing to the\nextreme workload of collecting data. However, most datasets rely on a rich\ncorpus of textual data allowing further mining of further information connected\nto each event. This paper proposes one such approach that is inexpensive and\nhigh performance, leveraging active learning - an iterative process of\nimproving a machine learning model based on sequential (guided) human input.\nActive learning is employed to then step-wise train (fine-tuning) of a large,\nencoder-only language model adapted for extracting sub-classes of events\nrelating to conflict dynamics. The approach shows performance similar to human\n(gold-standard) coding while reducing the amount of required human annotation\nby as much as 99%.\n", "date": "2024-02-02", "categories": []}
{"id": "2305.03923", "title": "Active Continual Learning: On Balancing Knowledge Retention and\n  Learnability", "abstract": "  Acquiring new knowledge without forgetting what has been learned in a\nsequence of tasks is the central focus of continual learning (CL). While tasks\narrive sequentially, the training data are often prepared and annotated\nindependently, leading to the CL of incoming supervised learning tasks. This\npaper considers the under-explored problem of active continual learning (ACL)\nfor a sequence of active learning (AL) tasks, where each incoming task includes\na pool of unlabelled data and an annotation budget. We investigate the\neffectiveness and interplay between several AL and CL algorithms in the domain,\nclass and task-incremental scenarios. Our experiments reveal the trade-off\nbetween two contrasting goals of not forgetting the old knowledge and the\nability to quickly learn new knowledge in CL and AL, respectively. While\nconditioning the AL query strategy on the annotations collected for the\nprevious tasks leads to improved task performance on the domain and task\nincremental learning, our proposed forgetting-learning profile suggests a gap\nin balancing the effect of AL and CL for the class-incremental scenario.\n", "date": "2023-05-06", "categories": []}
{"id": "2401.16634", "title": "The Why, When, and How to Use Active Learning in Large-Data-Driven 3D\n  Object Detection for Safe Autonomous Driving: An Empirical Exploration", "abstract": "  Active learning strategies for 3D object detection in autonomous driving\ndatasets may help to address challenges of data imbalance, redundancy, and\nhigh-dimensional data. We demonstrate the effectiveness of entropy querying to\nselect informative samples, aiming to reduce annotation costs and improve model\nperformance. We experiment using the BEVFusion model for 3D object detection on\nthe nuScenes dataset, comparing active learning to random sampling and\ndemonstrating that entropy querying outperforms in most cases. The method is\nparticularly effective in reducing the performance gap between majority and\nminority classes. Class-specific analysis reveals efficient allocation of\nannotated resources for limited data budgets, emphasizing the importance of\nselecting diverse and informative data for model training. Our findings suggest\nthat entropy querying is a promising strategy for selecting data that enhances\nmodel learning in resource-constrained environments.\n", "date": "2024-01-30", "categories": []}
{"id": "2401.16298", "title": "Breaking the Barrier: Selective Uncertainty-based Active Learning for\n  Medical Image Segmentation", "abstract": "  Active learning (AL) has found wide applications in medical image\nsegmentation, aiming to alleviate the annotation workload and enhance\nperformance. Conventional uncertainty-based AL methods, such as entropy and\nBayesian, often rely on an aggregate of all pixel-level metrics. However, in\nimbalanced settings, these methods tend to neglect the significance of target\nregions, eg., lesions, and tumors. Moreover, uncertainty-based selection\nintroduces redundancy. These factors lead to unsatisfactory performance, and in\nmany cases, even underperform random sampling. To solve this problem, we\nintroduce a novel approach called the Selective Uncertainty-based AL, avoiding\nthe conventional practice of summing up the metrics of all pixels. Through a\nfiltering process, our strategy prioritizes pixels within target areas and\nthose near decision boundaries. This resolves the aforementioned disregard for\ntarget areas and redundancy. Our method showed substantial improvements across\nfive different uncertainty-based methods and two distinct datasets, utilizing\nfewer labeled data to reach the supervised baseline and consistently achieving\nthe highest overall performance. Our code is available at\nhttps://github.com/HelenMa9998/Selective\\_Uncertainty\\_AL.\n", "date": "2024-01-29", "categories": []}
{"id": "2401.12722", "title": "Falcon: Fair Active Learning using Multi-armed Bandits", "abstract": "  Biased data can lead to unfair machine learning models, highlighting the\nimportance of embedding fairness at the beginning of data analysis,\nparticularly during dataset curation and labeling. In response, we propose\nFalcon, a scalable fair active learning framework. Falcon adopts a data-centric\napproach that improves machine learning model fairness via strategic sample\nselection. Given a user-specified group fairness measure, Falcon identifies\nsamples from \"target groups\" (e.g., (attribute=female, label=positive)) that\nare the most informative for improving fairness. However, a challenge arises\nsince these target groups are defined using ground truth labels that are not\navailable during sample selection. To handle this, we propose a novel\ntrial-and-error method, where we postpone using a sample if the predicted label\nis different from the expected one and falls outside the target group. We also\nobserve the trade-off that selecting more informative samples results in higher\nlikelihood of postponing due to undesired label prediction, and the optimal\nbalance varies per dataset. We capture the trade-off between informativeness\nand postpone rate as policies and propose to automatically select the best\npolicy using adversarial multi-armed bandit methods, given their computational\nefficiency and theoretical guarantees. Experiments show that Falcon\nsignificantly outperforms existing fair active learning approaches in terms of\nfairness and accuracy and is more efficient. In particular, only Falcon\nsupports a proper trade-off between accuracy and fairness where its maximum\nfairness score is 1.8-4.5x higher than the second-best results.\n", "date": "2024-01-23", "categories": []}
{"id": "2401.13229", "title": "From Random to Informed Data Selection: A Diversity-Based Approach to\n  Optimize Human Annotation and Few-Shot Learning", "abstract": "  A major challenge in Natural Language Processing is obtaining annotated data\nfor supervised learning. An option is the use of crowdsourcing platforms for\ndata annotation. However, crowdsourcing introduces issues related to the\nannotator's experience, consistency, and biases. An alternative is to use\nzero-shot methods, which in turn have limitations compared to their few-shot or\nfully supervised counterparts. Recent advancements driven by large language\nmodels show potential, but struggle to adapt to specialized domains with\nseverely limited data. The most common approaches therefore involve the human\nitself randomly annotating a set of datapoints to build initial datasets. But\nrandomly sampling data to be annotated is often inefficient as it ignores the\ncharacteristics of the data and the specific needs of the model. The situation\nworsens when working with imbalanced datasets, as random sampling tends to\nheavily bias towards the majority classes, leading to excessive annotated data.\nTo address these issues, this paper contributes an automatic and informed data\nselection architecture to build a small dataset for few-shot learning. Our\nproposal minimizes the quantity and maximizes diversity of data selected for\nhuman annotation, while improving model performance.\n", "date": "2024-01-24", "categories": []}
{"id": "2401.09555", "title": "Improving Classification Performance With Human Feedback: Label a few,\n  we label the rest", "abstract": "  In the realm of artificial intelligence, where a vast majority of data is\nunstructured, obtaining substantial amounts of labeled data to train supervised\nmachine learning models poses a significant challenge. To address this, we\ndelve into few-shot and active learning, where are goal is to improve AI models\nwith human feedback on a few labeled examples. This paper focuses on\nunderstanding how a continuous feedback loop can refine models, thereby\nenhancing their accuracy, recall, and precision through incremental human\ninput. By employing Large Language Models (LLMs) such as GPT-3.5, BERT, and\nSetFit, we aim to analyze the efficacy of using a limited number of labeled\nexamples to substantially improve model accuracy. We benchmark this approach on\nthe Financial Phrasebank, Banking, Craigslist, Trec, Amazon Reviews datasets to\nprove that with just a few labeled examples, we are able to surpass the\naccuracy of zero shot large language models to provide enhanced text\nclassification performance. We demonstrate that rather than needing to manually\nlabel millions of rows of data, we just need to label a few and the model can\neffectively predict the rest.\n", "date": "2024-01-17", "categories": []}
{"id": "2401.07367", "title": "Active Learning for NLP with Large Language Models", "abstract": "  Human annotation of training samples is expensive, laborious, and sometimes\nchallenging, especially for Natural Language Processing (NLP) tasks. To reduce\nthe labeling cost and enhance the sample efficiency, Active Learning (AL)\ntechnique can be used to label as few samples as possible to reach a reasonable\nor similar results. To reduce even more costs and with the significant advances\nof Large Language Models (LLMs), LLMs can be a good candidate to annotate\nsamples. This work investigates the accuracy and cost of using LLMs (GPT-3.5\nand GPT-4) to label samples on 3 different datasets. A consistency-based\nstrategy is proposed to select samples that are potentially incorrectly labeled\nso that human annotations can be used for those samples in AL settings, and we\ncall it mixed annotation strategy. Then we test performance of AL under two\ndifferent settings: (1) using human annotations only; (2) using the proposed\nmixed annotation strategy. The accuracy of AL models under 3 AL query\nstrategies are reported on 3 text classification datasets, i.e., AG's News,\nTREC-6, and Rotten Tomatoes. On AG's News and Rotten Tomatoes, the models\ntrained with the mixed annotation strategy achieves similar or better results\ncompared to that with human annotations. The method reveals great potentials of\nLLMs as annotators in terms of accuracy and cost efficiency in active learning\nsettings.\n", "date": "2024-01-14", "categories": []}
{"id": "2401.07395", "title": "Harnessing the Power of Beta Scoring in Deep Active Learning for\n  Multi-Label Text Classification", "abstract": "  Within the scope of natural language processing, the domain of multi-label\ntext classification is uniquely challenging due to its expansive and uneven\nlabel distribution. The complexity deepens due to the demand for an extensive\nset of annotated data for training an advanced deep learning model, especially\nin specialized fields where the labeling task can be labor-intensive and often\nrequires domain-specific knowledge. Addressing these challenges, our study\nintroduces a novel deep active learning strategy, capitalizing on the Beta\nfamily of proper scoring rules within the Expected Loss Reduction framework. It\ncomputes the expected increase in scores using the Beta Scoring Rules, which\nare then transformed into sample vector representations. These vector\nrepresentations guide the diverse selection of informative samples, directly\nlinking this process to the model's expected proper score. Comprehensive\nevaluations across both synthetic and real datasets reveal our method's\ncapability to often outperform established acquisition techniques in\nmulti-label text classification, presenting encouraging outcomes across various\narchitectural and dataset scenarios.\n", "date": "2024-01-15", "categories": []}
{"id": "2401.07639", "title": "Compute-Efficient Active Learning", "abstract": "  Active learning, a powerful paradigm in machine learning, aims at reducing\nlabeling costs by selecting the most informative samples from an unlabeled\ndataset. However, the traditional active learning process often demands\nextensive computational resources, hindering scalability and efficiency. In\nthis paper, we address this critical issue by presenting a novel method\ndesigned to alleviate the computational burden associated with active learning\non massive datasets. To achieve this goal, we introduce a simple, yet effective\nmethod-agnostic framework that outlines how to strategically choose and\nannotate data points, optimizing the process for efficiency while maintaining\nmodel performance. Through case studies, we demonstrate the effectiveness of\nour proposed method in reducing computational costs while maintaining or, in\nsome cases, even surpassing baseline model outcomes. Code is available at\nhttps://github.com/aimotive/Compute-Efficient-Active-Learning.\n", "date": "2024-01-15", "categories": []}
{"id": "2401.05465", "title": "D3GU: Multi-Target Active Domain Adaptation via Enhancing Domain\n  Alignment", "abstract": "  Unsupervised domain adaptation (UDA) for image classification has made\nremarkable progress in transferring classification knowledge from a labeled\nsource domain to an unlabeled target domain, thanks to effective domain\nalignment techniques. Recently, in order to further improve performance on a\ntarget domain, many Single-Target Active Domain Adaptation (ST-ADA) methods\nhave been proposed to identify and annotate the salient and exemplar target\nsamples. However, it requires one model to be trained and deployed for each\ntarget domain and the domain label associated with each test sample. This\nlargely restricts its application in the ubiquitous scenarios with multiple\ntarget domains. Therefore, we propose a Multi-Target Active Domain Adaptation\n(MT-ADA) framework for image classification, named D3GU, to simultaneously\nalign different domains and actively select samples from them for annotation.\nThis is the first research effort in this field to our best knowledge. D3GU\napplies Decomposed Domain Discrimination (D3) during training to achieve both\nsource-target and target-target domain alignments. Then during active sampling,\na Gradient Utility (GU) score is designed to weight every unlabeled target\nimage by its contribution towards classification and domain alignment tasks,\nand is further combined with KMeans clustering to form GU-KMeans for diverse\nimage sampling. Extensive experiments on three benchmark datasets, Office31,\nOfficeHome, and DomainNet, have been conducted to validate consistently\nsuperior performance of D3GU for MT-ADA.\n", "date": "2024-01-10", "categories": []}
{"id": "2401.04923", "title": "Inconsistency-Based Data-Centric Active Open-Set Annotation", "abstract": "  Active learning is a commonly used approach that reduces the labeling effort\nrequired to train deep neural networks. However, the effectiveness of current\nactive learning methods is limited by their closed-world assumptions, which\nassume that all data in the unlabeled pool comes from a set of predefined known\nclasses. This assumption is often not valid in practical situations, as there\nmay be unknown classes in the unlabeled data, leading to the active open-set\nannotation problem. The presence of unknown classes in the data can\nsignificantly impact the performance of existing active learning methods due to\nthe uncertainty they introduce. To address this issue, we propose a novel\ndata-centric active learning method called NEAT that actively annotates\nopen-set data. NEAT is designed to label known classes data from a pool of both\nknown and unknown classes unlabeled data. It utilizes the clusterability of\nlabels to identify the known classes from the unlabeled pool and selects\ninformative samples from those classes based on a consistency criterion that\nmeasures inconsistencies between model predictions and local feature\ndistribution. Unlike the recently proposed learning-centric method for the same\nproblem, NEAT is much more computationally efficient and is a data-centric\nactive open-set annotation method. Our experiments demonstrate that NEAT\nachieves significantly better performance than state-of-the-art active learning\nmethods for active open-set annotation.\n", "date": "2024-01-10", "categories": []}
{"id": "2401.01690", "title": "Zero-shot Active Learning Using Self Supervised Learning", "abstract": "  Deep learning algorithms are often said to be data hungry. The performance of\nsuch algorithms generally improve as more and more annotated data is fed into\nthe model. While collecting unlabelled data is easier (as they can be scraped\neasily from the internet), annotating them is a tedious and expensive task.\nGiven a fixed budget available for data annotation, Active Learning helps\nselecting the best subset of data for annotation, such that the deep learning\nmodel when trained over that subset will have maximum generalization\nperformance under this budget. In this work, we aim to propose a new Active\nLearning approach which is model agnostic as well as one doesn't require an\niterative process. We aim to leverage self-supervised learnt features for the\ntask of Active Learning. The benefit of self-supervised learning, is that one\ncan get useful feature representation of the input data, without having any\nannotation.\n", "date": "2024-01-03", "categories": []}
{"id": "2306.03543", "title": "How to Select Which Active Learning Strategy is Best Suited for Your\n  Specific Problem and Budget", "abstract": "  In the domain of Active Learning (AL), a learner actively selects which\nunlabeled examples to seek labels from an oracle, while operating within\npredefined budget constraints. Importantly, it has been recently shown that\ndistinct query strategies are better suited for different conditions and\nbudgetary constraints. In practice, the determination of the most appropriate\nAL strategy for a given situation remains an open problem. To tackle this\nchallenge, we propose a practical derivative-based method that dynamically\nidentifies the best strategy for a given budget. Intuitive motivation for our\napproach is provided by the theoretical analysis of a simplified scenario. We\nthen introduce a method to dynamically select an AL strategy, which takes into\naccount the unique characteristics of the problem and the available budget.\nEmpirical results showcase the effectiveness of our approach across diverse\nbudgets and computer vision tasks.\n", "date": "2023-06-06", "categories": []}
{"id": "2312.16931", "title": "DeLR: Active Learning for Detection with Decoupled Localization and\n  Recognition Query", "abstract": "  Active learning has been demonstrated effective to reduce labeling cost,\nwhile most progress has been designed for image recognition, there still lacks\ninstance-level active learning for object detection. In this paper, we rethink\ntwo key components, i.e., localization and recognition, for object detection,\nand find that the correctness of them are highly related, therefore, it is not\nnecessary to annotate both boxes and classes if we are given pseudo annotations\nprovided with the trained model. Motivated by this, we propose an efficient\nquery strategy, termed as DeLR, that Decoupling the Localization and\nRecognition for active query. In this way, we are probably free of class\nannotations when the localization is correct, and able to assign the labeling\nbudget for more informative samples. There are two main differences in DeLR: 1)\nUnlike previous methods mostly focus on image-level annotations, where the\nqueried samples are selected and exhausted annotated. In DeLR, the query is\nbased on region-level, and we only annotate the object region that is queried;\n2) Instead of directly providing both localization and recognition annotations,\nwe separately query the two components, and thus reduce the recognition budget\nwith the pseudo class labels provided by the model. Experiments on several\nbenchmarks demonstrate its superiority. We hope our proposed query strategy\nwould shed light on researches in active learning in object detection.\n", "date": "2023-12-28", "categories": []}
{"id": "2312.16965", "title": "Reinforcement-based Display-size Selection for Frugal Satellite Image\n  Change Detection", "abstract": "  We introduce a novel interactive satellite image change detection algorithm\nbased on active learning. The proposed method is iterative and consists in\nfrugally probing the user (oracle) about the labels of the most critical\nimages, and according to the oracle's annotations, it updates change detection\nresults. First, we consider a probabilistic framework which assigns to each\nunlabeled sample a relevance measure modeling how critical is that sample when\ntraining change detection functions. We obtain these relevance measures by\nminimizing an objective function mixing diversity, representativity and\nuncertainty. These criteria when combined allow exploring different data modes\nand also refining change detections. Then, we further explore the potential of\nthis objective function, by considering a reinforcement learning approach that\nfinds the best combination of diversity, representativity and uncertainty as\nwell as display-sizes through active learning iterations, leading to better\ngeneralization as shown through experiments in interactive satellite image\nchange detection.\n", "date": "2023-12-28", "categories": []}
{"id": "2209.14199", "title": "Active Transfer Prototypical Network: An Efficient Labeling Algorithm\n  for Time-Series Data", "abstract": "  The paucity of labeled data is a typical challenge in the automotive\nindustry. Annotating time-series measurements requires solid domain knowledge\nand in-depth exploratory data analysis, which implies a high labeling effort.\nConventional Active Learning (AL) addresses this issue by actively querying the\nmost informative instances based on the estimated classification probability\nand retraining the model iteratively. However, the learning efficiency strongly\nrelies on the initial model, resulting in the trade-off between the size of the\ninitial dataset and the query number. This paper proposes a novel Few-Shot\nLearning (FSL)-based AL framework, which addresses the trade-off problem by\nincorporating a Prototypical Network (ProtoNet) in the AL iterations. The\nresults show an improvement, on the one hand, in the robustness to the initial\nmodel and, on the other hand, in the learning efficiency of the ProtoNet\nthrough the active selection of the support set in each iteration. This\nframework was validated on UCI HAR/HAPT dataset and a real-world braking\nmaneuver dataset. The learning performance significantly surpasses traditional\nAL algorithms on both datasets, achieving 90% classification accuracy with 10%\nand 5% labeling effort, respectively.\n", "date": "2022-09-28", "categories": []}
{"id": "2312.15944", "title": "BAL: Balancing Diversity and Novelty for Active Learning", "abstract": "  The objective of Active Learning is to strategically label a subset of the\ndataset to maximize performance within a predetermined labeling budget. In this\nstudy, we harness features acquired through self-supervised learning. We\nintroduce a straightforward yet potent metric, Cluster Distance Difference, to\nidentify diverse data. Subsequently, we introduce a novel framework, Balancing\nActive Learning (BAL), which constructs adaptive sub-pools to balance diverse\nand uncertain data. Our approach outperforms all established active learning\nmethods on widely recognized benchmarks by 1.20%. Moreover, we assess the\nefficacy of our proposed framework under extended settings, encompassing both\nlarger and smaller labeling budgets. Experimental results demonstrate that,\nwhen labeling 80% of the samples, the performance of the current SOTA method\ndeclines by 0.74%, whereas our proposed BAL achieves performance comparable to\nthe full dataset. Codes are available at https://github.com/JulietLJY/BAL.\n", "date": "2023-12-26", "categories": []}
{"id": "2312.14126", "title": "Entropic Open-set Active Learning", "abstract": "  Active Learning (AL) aims to enhance the performance of deep models by\nselecting the most informative samples for annotation from a pool of unlabeled\ndata. Despite impressive performance in closed-set settings, most AL methods\nfail in real-world scenarios where the unlabeled data contains unknown\ncategories. Recently, a few studies have attempted to tackle the AL problem for\nthe open-set setting. However, these methods focus more on selecting known\nsamples and do not efficiently utilize unknown samples obtained during AL\nrounds. In this work, we propose an Entropic Open-set AL (EOAL) framework which\nleverages both known and unknown distributions effectively to select\ninformative samples during AL rounds. Specifically, our approach employs two\ndifferent entropy scores. One measures the uncertainty of a sample with respect\nto the known-class distributions. The other measures the uncertainty of the\nsample with respect to the unknown-class distributions. By utilizing these two\nentropy scores we effectively separate the known and unknown samples from the\nunlabeled data resulting in better sampling. Through extensive experiments, we\nshow that the proposed method outperforms existing state-of-the-art methods on\nCIFAR-10, CIFAR-100, and TinyImageNet datasets. Code is available at\n\\url{https://github.com/bardisafa/EOAL}.\n", "date": "2023-12-21", "categories": []}
{"id": "2312.12574", "title": "Generator Assisted Mixture of Experts For Feature Acquisition in Batch", "abstract": "  Given a set of observations, feature acquisition is about finding the subset\nof unobserved features which would enhance accuracy. Such problems have been\nexplored in a sequential setting in prior work. Here, the model receives\nfeedback from every new feature acquired and chooses to explore more features\nor to predict. However, sequential acquisition is not feasible in some settings\nwhere time is of the essence. We consider the problem of feature acquisition in\nbatch, where the subset of features to be queried in batch is chosen based on\nthe currently observed features, and then acquired as a batch, followed by\nprediction. We solve this problem using several technical innovations. First,\nwe use a feature generator to draw a subset of the synthetic features for some\nexamples, which reduces the cost of oracle queries. Second, to make the feature\nacquisition problem tractable for the large heterogeneous observed features, we\npartition the data into buckets, by borrowing tools from locality sensitive\nhashing and then train a mixture of experts model. Third, we design a tractable\nlower bound of the original objective. We use a greedy algorithm combined with\nmodel training to solve the underlying problem. Experiments with four datasets\nshow that our approach outperforms these methods in terms of trade-off between\naccuracy and feature acquisition cost.\n", "date": "2023-12-19", "categories": []}
{"id": "2209.12756", "title": "FAL-CUR: Fair Active Learning using Uncertainty and Representativeness\n  on Fair Clustering", "abstract": "  Active Learning (AL) techniques have proven to be highly effective in\nreducing data labeling costs across a range of machine learning tasks.\nNevertheless, one known challenge of these methods is their potential to\nintroduce unfairness towards sensitive attributes. Although recent approaches\nhave focused on enhancing fairness in AL, they tend to reduce the model's\naccuracy. To address this issue, we propose a novel strategy, named Fair Active\nLearning using fair Clustering, Uncertainty, and Representativeness (FAL-CUR),\nto improve fairness in AL. FAL-CUR tackles the fairness problem in AL by\ncombining fair clustering with an acquisition function that determines which\nsamples to query based on their uncertainty and representativeness scores. We\nevaluate the performance of FAL-CUR on four real-world datasets, and the\nresults demonstrate that FAL-CUR achieves a 15% - 20% improvement in fairness\ncompared to the best state-of-the-art method in terms of equalized odds while\nmaintaining stable accuracy scores. Furthermore, an ablation study highlights\nthe crucial roles of fair clustering in preserving fairness and the acquisition\nfunction in stabilizing the accuracy performance.\n", "date": "2022-09-21", "categories": []}
{"id": "2301.02761", "title": "Active Learning Guided by Efficient Surrogate Learners", "abstract": "  Re-training a deep learning model each time a single data point receives a\nnew label is impractical due to the inherent complexity of the training\nprocess. Consequently, existing active learning (AL) algorithms tend to adopt a\nbatch-based approach where, during each AL iteration, a set of data points is\ncollectively chosen for annotation. However, this strategy frequently leads to\nredundant sampling, ultimately eroding the efficacy of the labeling procedure.\nIn this paper, we introduce a new AL algorithm that harnesses the power of a\nGaussian process surrogate in conjunction with the neural network principal\nlearner. Our proposed model adeptly updates the surrogate learner for every new\ndata instance, enabling it to emulate and capitalize on the continuous learning\ndynamics of the neural network without necessitating a complete retraining of\nthe principal model for each individual label. Experiments on four benchmark\ndatasets demonstrate that this approach yields significant enhancements, either\nrivaling or aligning with the performance of state-of-the-art techniques.\n", "date": "2023-01-07", "categories": []}
{"id": "2302.07317", "title": "Algorithm Selection for Deep Active Learning with Imbalanced Datasets", "abstract": "  Label efficiency has become an increasingly important objective in deep\nlearning applications. Active learning aims to reduce the number of labeled\nexamples needed to train deep networks, but the empirical performance of active\nlearning algorithms can vary dramatically across datasets and applications. It\nis difficult to know in advance which active learning strategy will perform\nwell or best in a given application. To address this, we propose the first\nadaptive algorithm selection strategy for deep active learning. For any\nunlabeled dataset, our (meta) algorithm TAILOR (Thompson ActIve Learning\nalgORithm selection) iteratively and adaptively chooses among a set of\ncandidate active learning algorithms. TAILOR uses novel reward functions aimed\nat gathering class-balanced examples. Extensive experiments in multi-class and\nmulti-label applications demonstrate TAILOR's effectiveness in achieving\naccuracy comparable or better than that of the best of the candidate\nalgorithms. Our implementation of TAILOR is open-sourced at\nhttps://github.com/jifanz/TAILOR.\n", "date": "2023-02-14", "categories": []}
{"id": "2312.10116", "title": "Bayesian Estimate of Mean Proper Scores for Diversity-Enhanced Active\n  Learning", "abstract": "  The effectiveness of active learning largely depends on the sampling\nefficiency of the acquisition function. Expected Loss Reduction (ELR) focuses\non a Bayesian estimate of the reduction in classification error, and more\ngeneral costs fit in the same framework. We propose Bayesian Estimate of Mean\nProper Scores (BEMPS) to estimate the increase in strictly proper scores such\nas log probability or negative mean square error within this framework. We also\nprove convergence results for this general class of costs. To facilitate better\nexperimentation with the new acquisition functions, we develop a complementary\nbatch AL algorithm that encourages diversity in the vector of expected changes\nin scores for unlabeled data. To allow high-performance classifiers, we combine\ndeep ensembles, and dynamic validation set construction on pretrained models,\nand further speed up the ensemble process with the idea of Monte Carlo Dropout.\nExtensive experiments on both texts and images show that the use of mean square\nerror and log probability with BEMPS yields robust acquisition functions and\nwell-calibrated classifiers, and consistently outperforms the others tested.\nThe advantages of BEMPS over the others are further supported by a set of\nqualitative analyses, where we visualise their sampling behaviour using data\nmaps and t-SNE plots.\n", "date": "2023-12-15", "categories": []}
{"id": "2312.10817", "title": "Ocean Data Quality Assessment through Outlier Detection-enhanced Active\n  Learning", "abstract": "  Ocean and climate research benefits from global ocean observation initiatives\nsuch as Argo, GLOSS, and EMSO. The Argo network, dedicated to ocean profiling,\ngenerates a vast volume of observatory data. However, data quality issues from\nsensor malfunctions and transmission errors necessitate stringent quality\nassessment. Existing methods, including machine learning, fall short due to\nlimited labeled data and imbalanced datasets. To address these challenges, we\npropose an ODEAL framework for ocean data quality assessment, employing AL to\nreduce human experts' workload in the quality assessment workflow and\nleveraging outlier detection algorithms for effective model initialization. We\nalso conduct extensive experiments on five large-scale realistic Argo datasets\nto gain insights into our proposed method, including the effectiveness of AL\nquery strategies and the initial set construction approach. The results suggest\nthat our framework enhances quality assessment efficiency by up to 465.5% with\nthe uncertainty-based query strategy compared to random sampling and minimizes\noverall annotation costs by up to 76.9% using the initial set built with\noutlier detectors.\n", "date": "2023-12-17", "categories": []}
{"id": "2309.01013", "title": "Streaming Active Learning for Regression Problems Using Regression via\n  Classification", "abstract": "  One of the challenges in deploying a machine learning model is that the\nmodel's performance degrades as the operating environment changes. To maintain\nthe performance, streaming active learning is used, in which the model is\nretrained by adding a newly annotated sample to the training dataset if the\nprediction of the sample is not certain enough. Although many streaming active\nlearning methods have been proposed for classification, few efforts have been\nmade for regression problems, which are often handled in the industrial field.\nIn this paper, we propose to use the regression-via-classification framework\nfor streaming active learning for regression. Regression-via-classification\ntransforms regression problems into classification problems so that streaming\nactive learning methods proposed for classification problems can be applied\ndirectly to regression problems. Experimental validation on four real data sets\nshows that the proposed method can perform regression with higher accuracy at\nthe same annotation cost.\n", "date": "2023-09-02", "categories": []}
{"id": "2312.08559", "title": "Fair Active Learning in Low-Data Regimes", "abstract": "  In critical machine learning applications, ensuring fairness is essential to\navoid perpetuating social inequities. In this work, we address the challenges\nof reducing bias and improving accuracy in data-scarce environments, where the\ncost of collecting labeled data prohibits the use of large, labeled datasets.\nIn such settings, active learning promises to maximize marginal accuracy gains\nof small amounts of labeled data. However, existing applications of active\nlearning for fairness fail to deliver on this, typically requiring large\nlabeled datasets, or failing to ensure the desired fairness tolerance is met on\nthe population distribution.\n  To address such limitations, we introduce an innovative active learning\nframework that combines an exploration procedure inspired by posterior sampling\nwith a fair classification subroutine. We demonstrate that this framework\nperforms effectively in very data-scarce regimes, maximizing accuracy while\nsatisfying fairness constraints with high probability. We evaluate our proposed\napproach using well-established real-world benchmark datasets and compare it\nagainst state-of-the-art methods, demonstrating its effectiveness in producing\nfair models, and improvement over existing methods.\n", "date": "2023-12-13", "categories": []}
{"id": "2305.06408", "title": "Accelerating Batch Active Learning Using Continual Learning Techniques", "abstract": "  A major problem with Active Learning (AL) is high training costs since models\nare typically retrained from scratch after every query round. We start by\ndemonstrating that standard AL on neural networks with warm starting fails,\nboth to accelerate training and to avoid catastrophic forgetting when using\nfine-tuning over AL query rounds. We then develop a new class of techniques,\ncircumventing this problem, by biasing further training towards previously\nlabeled sets. We accomplish this by employing existing, and developing novel,\nreplay-based Continual Learning (CL) algorithms that are effective at quickly\nlearning the new without forgetting the old, especially when data comes from an\nevolving distribution. We call this paradigm Continual Active Learning (CAL).\nWe show CAL achieves significant speedups using a plethora of replay schemes\nthat use model distillation and that select diverse, uncertain points from the\nhistory. We conduct experiments across many data domains, including natural\nlanguage, vision, medical imaging, and computational biology, each with\ndifferent neural architectures and dataset sizes. CAL consistently provides a\n3x reduction in training time, while retaining performance.\n", "date": "2023-05-10", "categories": []}
{"id": "2312.08198", "title": "Towards Model-Based Data Acquisition for Subjective Multi-Task NLP\n  Problems", "abstract": "  Data annotated by humans is a source of knowledge by describing the\npeculiarities of the problem and therefore fueling the decision process of the\ntrained model. Unfortunately, the annotation process for subjective natural\nlanguage processing (NLP) problems like offensiveness or emotion detection is\noften very expensive and time-consuming. One of the inevitable risks is to\nspend some of the funds and annotator effort on annotations that do not provide\nany additional knowledge about the specific task. To minimize these costs, we\npropose a new model-based approach that allows the selection of tasks annotated\nindividually for each text in a multi-task scenario. The experiments carried\nout on three datasets, dozens of NLP tasks, and thousands of annotations show\nthat our method allows up to 40% reduction in the number of annotations with\nnegligible loss of knowledge. The results also emphasize the need to collect a\ndiverse amount of data required to efficiently train a model, depending on the\nsubjectivity of the annotation task. We also focused on measuring the relation\nbetween subjective tasks by evaluating the model in single-task and multi-task\nscenarios. Moreover, for some datasets, training only on the labels predicted\nby our model improved the efficiency of task selection as a self-supervised\nlearning regularization technique.\n", "date": "2023-12-13", "categories": []}
{"id": "2312.05751", "title": "Benchmarking of Query Strategies: Towards Future Deep Active Learning", "abstract": "  In this study, we benchmark query strategies for deep actice learning~(DAL).\nDAL reduces annotation costs by annotating only high-quality samples selected\nby query strategies. Existing research has two main problems, that the\nexperimental settings are not standardized, making the evaluation of existing\nmethods is difficult, and that most of experiments were conducted on the CIFAR\nor MNIST datasets. Therefore, we develop standardized experimental settings for\nDAL and investigate the effectiveness of various query strategies using six\ndatasets, including those that contain medical and visual inspection images. In\naddition, since most current DAL approaches are model-based, we perform\nverification experiments using fully-trained models for querying to investigate\nthe effectiveness of these approaches for the six datasets. Our code is\navailable at\n\\href{https://github.com/ia-gu/Benchmarking-of-Query-Strategies-Towards-Future-Deep-Active-Learning}\n", "date": "2023-12-10", "categories": []}
{"id": "2306.00945", "title": "CS4ML: A general framework for active learning with arbitrary data based\n  on Christoffel functions", "abstract": "  We introduce a general framework for active learning in regression problems.\nOur framework extends the standard setup by allowing for general types of data,\nrather than merely pointwise samples of the target function. This\ngeneralization covers many cases of practical interest, such as data acquired\nin transform domains (e.g., Fourier data), vector-valued data (e.g.,\ngradient-augmented data), data acquired along continuous curves, and,\nmultimodal data (i.e., combinations of different types of measurements). Our\nframework considers random sampling according to a finite number of sampling\nmeasures and arbitrary nonlinear approximation spaces (model classes). We\nintroduce the concept of generalized Christoffel functions and show how these\ncan be used to optimize the sampling measures. We prove that this leads to\nnear-optimal sample complexity in various important cases. This paper focuses\non applications in scientific computing, where active learning is often\ndesirable, since it is usually expensive to generate data. We demonstrate the\nefficacy of our framework for gradient-augmented learning with polynomials,\nMagnetic Resonance Imaging (MRI) using generative models and adaptive sampling\nfor solving PDEs using Physics-Informed Neural Networks (PINNs).\n", "date": "2023-06-01", "categories": []}
{"id": "2312.04604", "title": "Transferable Candidate Proposal with Bounded Uncertainty", "abstract": "  From an empirical perspective, the subset chosen through active learning\ncannot guarantee an advantage over random sampling when transferred to another\nmodel. While it underscores the significance of verifying transferability,\nexperimental design from previous works often neglected that the\ninformativeness of a data subset can change over model configurations. To\ntackle this issue, we introduce a new experimental design, coined as Candidate\nProposal, to find transferable data candidates from which active learning\nalgorithms choose the informative subset. Correspondingly, a data selection\nalgorithm is proposed, namely Transferable candidate proposal with Bounded\nUncertainty (TBU), which constrains the pool of transferable data candidates by\nfiltering out the presumably redundant data points based on uncertainty\nestimation. We verified the validity of TBU in image classification benchmarks,\nincluding CIFAR-10/100 and SVHN. When transferred to different model\nconfigurations, TBU consistency improves performance in existing active\nlearning algorithms. Our code is available at\nhttps://github.com/gokyeongryeol/TBU.\n", "date": "2023-12-07", "categories": []}
{"id": "2312.03619", "title": "Evaluation of Active Feature Acquisition Methods for Static Feature\n  Settings", "abstract": "  Active feature acquisition (AFA) agents, crucial in domains like healthcare\nwhere acquiring features is often costly or harmful, determine the optimal set\nof features for a subsequent classification task. As deploying an AFA agent\nintroduces a shift in missingness distribution, it's vital to assess its\nexpected performance at deployment using retrospective data. In a companion\npaper, we introduce a semi-offline reinforcement learning (RL) framework for\nactive feature acquisition performance evaluation (AFAPE) where features are\nassumed to be time-dependent. Here, we study and extend the AFAPE problem to\ncover static feature settings, where features are time-invariant, and hence\nprovide more flexibility to the AFA agents in deciding the order of the\nacquisitions. In this static feature setting, we derive and adapt new inverse\nprobability weighting (IPW), direct method (DM), and double reinforcement\nlearning (DRL) estimators within the semi-offline RL framework. These\nestimators can be applied when the missingness in the retrospective dataset\nfollows a missing-at-random (MAR) pattern. They also can be applied to\nmissing-not-at-random (MNAR) patterns in conjunction with appropriate existing\nmissing data techniques. We illustrate the improved data efficiency offered by\nthe semi-offline RL estimators in synthetic and real-world data experiments\nunder synthetic MAR and MNAR missingness.\n", "date": "2023-12-06", "categories": []}
{"id": "2312.04307", "title": "A Structural-Clustering Based Active Learning for Graph Neural Networks", "abstract": "  In active learning for graph-structured data, Graph Neural Networks (GNNs)\nhave shown effectiveness. However, a common challenge in these applications is\nthe underutilization of crucial structural information. To address this\nproblem, we propose the Structural-Clustering PageRank method for improved\nActive learning (SPA) specifically designed for graph-structured data. SPA\nintegrates community detection using the SCAN algorithm with the PageRank\nscoring method for efficient and informative sample selection. SPA prioritizes\nnodes that are not only informative but also central in structure. Through\nextensive experiments, SPA demonstrates higher accuracy and macro-F1 score over\nexisting methods across different annotation budgets and achieves significant\nreductions in query time. In addition, the proposed method only adds two\nhyperparameters, $\\epsilon$ and $\\mu$ in the algorithm to finely tune the\nbalance between structural learning and node selection. This simplicity is a\nkey advantage in active learning scenarios, where extensive hyperparameter\ntuning is often impractical.\n", "date": "2023-12-07", "categories": []}
{"id": "2312.03176", "title": "Active Learning for Abrupt Shifts Change-point Detection via\n  Derivative-Aware Gaussian Processes", "abstract": "  Change-point detection (CPD) is crucial for identifying abrupt shifts in\ndata, which influence decision-making and efficient resource allocation across\nvarious domains. To address the challenges posed by the costly and\ntime-intensive data acquisition in CPD, we introduce the Derivative-Aware\nChange Detection (DACD) method. It leverages the derivative process of a\nGaussian process (GP) for Active Learning (AL), aiming to pinpoint change-point\nlocations effectively. DACD balances the exploitation and exploration of\nderivative processes through multiple data acquisition functions (AFs). By\nutilizing GP derivative mean and variance as criteria, DACD sequentially\nselects the next sampling data point, thus enhancing algorithmic efficiency and\nensuring reliable and accurate results. We investigate the effectiveness of\nDACD method in diverse scenarios and show it outperforms other active learning\nchange-point detection approaches.\n", "date": "2023-12-05", "categories": []}
{"id": "2312.02247", "title": "Federated Active Learning for Target Domain Generalisation", "abstract": "  In this paper, we introduce Active Learning framework in Federated Learning\nfor Target Domain Generalisation, harnessing the strength from both learning\nparadigms. Our framework, FEDALV, composed of Active Learning (AL) and\nFederated Domain Generalisation (FDG), enables generalisation of an image\nclassification model trained from limited source domain client's data without\nsharing images to an unseen target domain. To this end, our FDG, FEDA, consists\nof two optimisation updates during training, one at the client and another at\nthe server level. For the client, the introduced losses aim to reduce feature\ncomplexity and condition alignment, while in the server, the regularisation\nlimits free energy biases between source and target obtained by the global\nmodel. The remaining component of FEDAL is AL with variable budgets, which\nqueries the server to retrieve and sample the most informative local data for\nthe targeted client. We performed multiple experiments on FDG w/ and w/o AL and\ncompared with both conventional FDG baselines and Federated Active Learning\nbaselines. Our extensive quantitative experiments demonstrate the superiority\nof our method in accuracy and efficiency compared to the multiple contemporary\nmethods. FEDALV manages to obtain the performance of the full training target\naccuracy while sampling as little as 5% of the source client's data.\n", "date": "2023-12-04", "categories": []}
{"id": "2312.01588", "title": "ActiveClean: Generating Line-Level Vulnerability Data via Active\n  Learning", "abstract": "  Deep learning vulnerability detection tools are increasing in popularity and\nhave been shown to be effective. These tools rely on large volume of high\nquality training data, which are very hard to get. Most of the currently\navailable datasets provide function-level labels, reporting whether a function\nis vulnerable or not vulnerable. However, for a vulnerability detection to be\nuseful, we need to also know the lines that are relevant to the vulnerability.\nThis paper makes efforts towards developing systematic tools and proposes.\nActiveClean to generate the large volume of line-level vulnerability data from\ncommits. That is, in addition to function-level labels, it also reports which\nlines in the function are likely responsible for vulnerability detection. In\nthe past, static analysis has been applied to clean commits to generate\nline-level data. Our approach based on active learning, which is easy to use\nand scalable, provide a complementary approach to static analysis. We designed\nsemantic and syntactic properties from commit lines and use them to train the\nmodel. We evaluated our approach on both Java and C datasets processing more\nthan 4.3K commits and 119K commit lines. AcitveClean achieved an F1 score\nbetween 70-74. Further, we also show that active learning is effective by using\njust 400 training data to reach F1 score of 70.23. Using ActiveClean, we\ngenerate the line-level labels for the entire FFMpeg project in the Devign\ndataset, including 5K functions, and also detected incorrect function-level\nlabels. We demonstrated that using our cleaned data, LineVul, a SOTA line-level\nvulnerability detection tool, detected 70 more vulnerable lines and 18 more\nvulnerable functions, and improved Top 10 accuracy from 66% to 73%.\n", "date": "2023-12-04", "categories": []}
{"id": "2312.00364", "title": "Benchmarking Multi-Domain Active Learning on Image Classification", "abstract": "  Active learning aims to enhance model performance by strategically labeling\ninformative data points. While extensively studied, its effectiveness on\nlarge-scale, real-world datasets remains underexplored. Existing research\nprimarily focuses on single-source data, ignoring the multi-domain nature of\nreal-world data. We introduce a multi-domain active learning benchmark to\nbridge this gap. Our benchmark demonstrates that traditional single-domain\nactive learning strategies are often less effective than random selection in\nmulti-domain scenarios. We also introduce CLIP-GeoYFCC, a novel large-scale\nimage dataset built around geographical domains, in contrast to existing\ngenre-based domain datasets. Analysis on our benchmark shows that all\nmulti-domain strategies exhibit significant tradeoffs, with no strategy\noutperforming across all datasets or all metrics, emphasizing the need for\nfuture research.\n", "date": "2023-12-01", "categories": []}
{"id": "2302.08893", "title": "Active learning for data streams: a survey", "abstract": "  Online active learning is a paradigm in machine learning that aims to select\nthe most informative data points to label from a data stream. The problem of\nminimizing the cost associated with collecting labeled observations has gained\na lot of attention in recent years, particularly in real-world applications\nwhere data is only available in an unlabeled form. Annotating each observation\ncan be time-consuming and costly, making it difficult to obtain large amounts\nof labeled data. To overcome this issue, many active learning strategies have\nbeen proposed in the last decades, aiming to select the most informative\nobservations for labeling in order to improve the performance of machine\nlearning models. These approaches can be broadly divided into two categories:\nstatic pool-based and stream-based active learning. Pool-based active learning\ninvolves selecting a subset of observations from a closed pool of unlabeled\ndata, and it has been the focus of many surveys and literature reviews.\nHowever, the growing availability of data streams has led to an increase in the\nnumber of approaches that focus on online active learning, which involves\ncontinuously selecting and labeling observations as they arrive in a stream.\nThis work aims to provide an overview of the most recently proposed approaches\nfor selecting the most informative observations from data streams in real time.\nWe review the various techniques that have been proposed and discuss their\nstrengths and limitations, as well as the challenges and opportunities that\nexist in this area of research.\n", "date": "2023-02-17", "categories": []}
{"id": "2311.15516", "title": "Active Foundational Models for Fault Diagnosis of Electrical Motors", "abstract": "  Fault detection and diagnosis of electrical motors are of utmost importance\nin ensuring the safe and reliable operation of several industrial systems.\nDetection and diagnosis of faults at the incipient stage allows corrective\nactions to be taken in order to reduce the severity of faults. The existing\ndata-driven deep learning approaches for machine fault diagnosis rely\nextensively on huge amounts of labeled samples, where annotations are expensive\nand time-consuming. However, a major portion of unlabeled condition monitoring\ndata is not exploited in the training process. To overcome this limitation, we\npropose a foundational model-based Active Learning framework that utilizes less\namount of labeled samples, which are most informative and harnesses a large\namount of available unlabeled data by effectively combining Active Learning and\nContrastive Self-Supervised Learning techniques. It consists of a transformer\nnetwork-based backbone model trained using an advanced nearest-neighbor\ncontrastive self-supervised learning method. This approach empowers the\nbackbone to learn improved representations of samples derived from raw,\nunlabeled vibration data. Subsequently, the backbone can undergo fine-tuning to\naddress a range of downstream tasks, both within the same machines and across\ndifferent machines. The effectiveness of the proposed methodology has been\nassessed through the fine-tuning of the backbone for multiple target tasks\nusing three distinct machine-bearing fault datasets. The experimental\nevaluation demonstrates a superior performance as compared to existing\nstate-of-the-art fault diagnosis methods with less amount of labeled data.\n", "date": "2023-11-27", "categories": []}
{"id": "2311.15614", "title": "FreeAL: Towards Human-Free Active Learning in the Era of Large Language\n  Models", "abstract": "  Collecting high-quality labeled data for model training is notoriously\ntime-consuming and labor-intensive for various NLP tasks. While copious\nsolutions, such as active learning for small language models (SLMs) and\nprevalent in-context learning in the era of large language models (LLMs), have\nbeen proposed and alleviate the labeling burden to some extent, their\nperformances are still subject to human intervention. It is still underexplored\nhow to reduce the annotation cost in the LLMs era. To bridge this, we\nrevolutionize traditional active learning and propose an innovative\ncollaborative learning framework FreeAL to interactively distill and filter the\ntask-specific knowledge from LLMs. During collaborative training, an LLM serves\nas an active annotator inculcating its coarse-grained knowledge, while a\ndownstream SLM is incurred as a student to filter out high-quality in-context\nsamples to feedback LLM for the subsequent label refinery. Extensive\nexperiments on eight benchmark datasets demonstrate that FreeAL largely\nenhances the zero-shot performances for both SLM and LLM without any human\nsupervision. The code is available at https://github.com/Justherozen/FreeAL .\n", "date": "2023-11-27", "categories": []}
{"id": "2305.14264", "title": "Active Learning Principles for In-Context Learning with Large Language\n  Models", "abstract": "  The remarkable advancements in large language models (LLMs) have\nsignificantly enhanced the performance in few-shot learning settings. By using\nonly a small number of labeled examples, referred to as demonstrations, LLMs\ncan effectively grasp the task at hand through in-context learning. However,\nthe process of selecting appropriate demonstrations has received limited\nattention in prior work. This paper addresses the issue of identifying the most\ninformative demonstrations for few-shot learning by approaching it as a\npool-based Active Learning (AL) problem over a single iteration. Our objective\nis to investigate how AL algorithms can serve as effective demonstration\nselection methods for in-context learning. We compare various standard AL\nalgorithms based on uncertainty, diversity, and similarity, and consistently\nobserve that the latter outperforms all other methods, including random\nsampling. Notably, uncertainty sampling, despite its success in conventional\nsupervised learning scenarios, performs poorly in this context. Our extensive\nexperimentation involving a diverse range of GPT and OPT models across $24$\nclassification and multi-choice tasks, coupled with thorough analysis,\nunambiguously demonstrates that in-context example selection through AL\nprioritizes high-quality examples that exhibit low uncertainty and bear\nsimilarity to the test examples.\n", "date": "2023-05-23", "categories": []}
{"id": "2311.12905", "title": "Revisiting the Domain Shift and Sample Uncertainty in Multi-source\n  Active Domain Transfer", "abstract": "  Active Domain Adaptation (ADA) aims to maximally boost model adaptation in a\nnew target domain by actively selecting a limited number of target data to\nannotate.This setting neglects the more practical scenario where training data\nare collected from multiple sources. This motivates us to target a new and\nchallenging setting of knowledge transfer that extends ADA from a single source\ndomain to multiple source domains, termed Multi-source Active Domain Adaptation\n(MADA). Not surprisingly, we find that most traditional ADA methods cannot work\ndirectly in such a setting, mainly due to the excessive domain gap introduced\nby all the source domains and thus their uncertainty-aware sample selection can\neasily become miscalibrated under the multi-domain shifts. Considering this, we\npropose a Dynamic integrated uncertainty valuation framework(Detective) that\ncomprehensively consider the domain shift between multi-source domains and\ntarget domain to detect the informative target samples. Specifically, the\nleverages a dynamic Domain Adaptation(DA) model that learns how to adapt the\nmodel's parameters to fit the union of multi-source domains. This enables an\napproximate single-source domain modeling by the dynamic model. We then\ncomprehensively measure both domain uncertainty and predictive uncertainty in\nthe target domain to detect informative target samples using evidential deep\nlearning, thereby mitigating uncertainty miscalibration. Furthermore, we\nintroduce a contextual diversity-aware calculator to enhance the diversity of\nthe selected samples. Experiments demonstrate that our solution outperforms\nexisting methods by a considerable margin on three domain adaptation\nbenchmarks.\n", "date": "2023-11-21", "categories": []}
{"id": "2308.07121", "title": "Active Bird2Vec: Towards End-to-End Bird Sound Monitoring with\n  Transformers", "abstract": "  We propose a shift towards end-to-end learning in bird sound monitoring by\ncombining self-supervised (SSL) and deep active learning (DAL). Leveraging\ntransformer models, we aim to bypass traditional spectrogram conversions,\nenabling direct raw audio processing. ActiveBird2Vec is set to generate\nhigh-quality bird sound representations through SSL, potentially accelerating\nthe assessment of environmental changes and decision-making processes for wind\nfarms. Additionally, we seek to utilize the wide variety of bird vocalizations\nthrough DAL, reducing the reliance on extensively labeled datasets by human\nexperts. We plan to curate a comprehensive set of tasks through Huggingface\nDatasets, enhancing future comparability and reproducibility of bioacoustic\nresearch. A comparative analysis between various transformer models will be\nconducted to evaluate their proficiency in bird sound recognition tasks. We aim\nto accelerate the progression of avian bioacoustic research and contribute to\nmore effective conservation strategies.\n", "date": "2023-08-14", "categories": []}
{"id": "2311.10591", "title": "FOCAL: A Cost-Aware Video Dataset for Active Learning", "abstract": "  In this paper, we introduce the FOCAL (Ford-OLIVES Collaboration on Active\nLearning) dataset which enables the study of the impact of annotation-cost\nwithin a video active learning setting. Annotation-cost refers to the time it\ntakes an annotator to label and quality-assure a given video sequence. A\npractical motivation for active learning research is to minimize\nannotation-cost by selectively labeling informative samples that will maximize\nperformance within a given budget constraint. However, previous work in video\nactive learning lacks real-time annotation labels for accurately assessing cost\nminimization and instead operates under the assumption that annotation-cost\nscales linearly with the amount of data to annotate. This assumption does not\ntake into account a variety of real-world confounding factors that contribute\nto a nonlinear cost such as the effect of an assistive labeling tool and the\nvariety of interactions within a scene such as occluded objects, weather, and\nmotion of objects. FOCAL addresses this discrepancy by providing real\nannotation-cost labels for 126 video sequences across 69 unique city scenes\nwith a variety of weather, lighting, and seasonal conditions. We also introduce\na set of conformal active learning algorithms that take advantage of the\nsequential structure of video data in order to achieve a better trade-off\nbetween annotation-cost and performance while also reducing floating point\noperations (FLOPS) overhead by at least 77.67%. We show how these approaches\nbetter reflect how annotations on videos are done in practice through a\nsequence selection framework. We further demonstrate the advantage of these\napproaches by introducing two performance-cost metrics and show that the best\nconformal active learning method is cheaper than the best traditional active\nlearning method by 113 hours.\n", "date": "2023-11-17", "categories": []}
{"id": "2311.09825", "title": "Human Still Wins over LLM: An Empirical Study of Active Learning on\n  Domain-Specific Annotation Tasks", "abstract": "  Large Language Models (LLMs) have demonstrated considerable advances, and\nseveral claims have been made about their exceeding human performance. However,\nin real-world tasks, domain knowledge is often required. Low-resource learning\nmethods like Active Learning (AL) have been proposed to tackle the cost of\ndomain expert annotation, raising this question: Can LLMs surpass compact\nmodels trained with expert annotations in domain-specific tasks? In this work,\nwe conduct an empirical experiment on four datasets from three different\ndomains comparing SOTA LLMs with small models trained on expert annotations\nwith AL. We found that small models can outperform GPT-3.5 with a few hundreds\nof labeled data, and they achieve higher or similar performance with GPT-4\ndespite that they are hundreds time smaller. Based on these findings, we posit\nthat LLM predictions can be used as a warmup method in real-world applications\nand human experts remain indispensable in tasks involving data annotation\ndriven by domain-specific knowledge.\n", "date": "2023-11-16", "categories": []}
{"id": "2311.06379", "title": "DeMuX: Data-efficient Multilingual Learning", "abstract": "  We consider the task of optimally fine-tuning pre-trained multilingual\nmodels, given small amounts of unlabelled target data and an annotation budget.\nIn this paper, we introduce DEMUX, a framework that prescribes the exact\ndata-points to label from vast amounts of unlabelled multilingual data, having\nunknown degrees of overlap with the target set. Unlike most prior works, our\nend-to-end framework is language-agnostic, accounts for model representations,\nand supports multilingual target configurations. Our active learning strategies\nrely upon distance and uncertainty measures to select task-specific neighbors\nthat are most informative to label, given a model. DeMuX outperforms strong\nbaselines in 84% of the test cases, in the zero-shot setting of disjoint source\nand target language sets (including multilingual target pools), across three\nmodels and four tasks. Notably, in low-budget settings (5-100 examples), we\nobserve gains of up to 8-11 F1 points for token-level tasks, and 2-5 F1 for\ncomplex tasks. Our code is released here:\nhttps://github.com/simran-khanuja/demux.\n", "date": "2023-11-10", "categories": []}
{"id": "2307.14899", "title": "Retrieval-based Text Selection for Addressing Class-Imbalanced Data in\n  Classification", "abstract": "  This paper addresses the problem of selecting of a set of texts for\nannotation in text classification using retrieval methods when there are limits\non the number of annotations due to constraints on human resources. An\nadditional challenge addressed is dealing with binary categories that have a\nsmall number of positive instances, reflecting severe class imbalance. In our\nsituation, where annotation occurs over a long time period, the selection of\ntexts to be annotated can be made in batches, with previous annotations guiding\nthe choice of the next set. To address these challenges, the paper proposes\nleveraging SHAP to construct a quality set of queries for Elasticsearch and\nsemantic search, to try to identify optimal sets of texts for annotation that\nwill help with class imbalance. The approach is tested on sets of cue texts\ndescribing possible future events, constructed by participants involved in\nstudies aimed to help with the management of obesity and diabetes. We introduce\nan effective method for selecting a small set of texts for annotation and\nbuilding high-quality classifiers. We integrate vector search, semantic search,\nand machine learning classifiers to yield a good solution. Our experiments\ndemonstrate improved F1 scores for the minority classes in binary\nclassification.\n", "date": "2023-07-27", "categories": []}
{"id": "2311.06057", "title": "Ulcerative Colitis Mayo Endoscopic Scoring Classification with Active\n  Learning and Generative Data Augmentation", "abstract": "  Endoscopic imaging is commonly used to diagnose Ulcerative Colitis (UC) and\nclassify its severity. It has been shown that deep learning based methods are\neffective in automated analysis of these images and can potentially be used to\naid medical doctors. Unleashing the full potential of these methods depends on\nthe availability of large amount of labeled images; however, obtaining and\nlabeling these images are quite challenging. In this paper, we propose a active\nlearning based generative augmentation method. The method involves generating a\nlarge number of synthetic samples by training using a small dataset consisting\nof real endoscopic images. The resulting data pool is narrowed down by using\nactive learning methods to select the most informative samples, which are then\nused to train a classifier. We demonstrate the effectiveness of our method\nthrough experiments on a publicly available endoscopic image dataset. The\nresults show that using synthesized samples in conjunction with active learning\nleads to improved classification performance compared to using only the\noriginal labeled examples and the baseline classification performance of 68.1%\nincreases to 74.5% in terms of Quadratic Weighted Kappa (QWK) Score. Another\nobservation is that, attaining equivalent performance using only real data\nnecessitated three times higher number of images.\n", "date": "2023-11-10", "categories": []}
{"id": "2311.05501", "title": "Dirichlet Active Learning", "abstract": "  This work introduces Dirichlet Active Learning (DiAL), a Bayesian-inspired\napproach to the design of active learning algorithms. Our framework models\nfeature-conditional class probabilities as a Dirichlet random field and lends\nobservational strength between similar features in order to calibrate the\nrandom field. This random field can then be utilized in learning tasks: in\nparticular, we can use current estimates of mean and variance to conduct\nclassification and active learning in the context where labeled data is scarce.\nWe demonstrate the applicability of this model to low-label rate graph learning\nby constructing ``propagation operators'' based upon the graph Laplacian, and\noffer computational studies demonstrating the method's competitiveness with the\nstate of the art. Finally, we provide rigorous guarantees regarding the ability\nof this approach to ensure both exploration and exploitation, expressed\nrespectively in terms of cluster exploration and increased attention to\ndecision boundaries.\n", "date": "2023-11-09", "categories": []}
{"id": "2206.09798", "title": "Using Sum-Product Networks to Assess Uncertainty in Deep Active Learning", "abstract": "  The success of deep active learning hinges on the choice of an effective\nacquisition function, which ranks not yet labeled data points according to\ntheir expected informativeness. Many acquisition functions are (partly) based\non the uncertainty that the current model has about the class label of a point,\nyet there is no generally agreed upon strategy for computing such uncertainty.\nThis paper proposes a new and very simple approach to computing uncertainty in\ndeep active learning with a Convolutional Neural Network (CNN). The main idea\nis to use the feature representation extracted by the CNN as data for training\na Sum-Product Network (SPN). Since SPNs are typically used for estimating the\ndistribution of a dataset, they are well suited to the task of estimating class\nprobabilities that can be used directly by standard acquisition functions such\nas max entropy and variational ratio. The effectiveness of our method is\ndemonstrated in an experimental study on several standard benchmark datasets\nfor image classification, where we compare it to various state-of-the-art\nmethods for assessing uncertainty in deep active learning.\n", "date": "2022-06-20", "categories": []}
{"id": "2309.09319", "title": "Active Learning for Semantic Segmentation with Multi-class Label Query", "abstract": "  This paper proposes a new active learning method for semantic segmentation.\nThe core of our method lies in a new annotation query design. It samples\ninformative local image regions (e.g., superpixels), and for each of such\nregions, asks an oracle for a multi-hot vector indicating all classes existing\nin the region. This multi-class labeling strategy is substantially more\nefficient than existing ones like segmentation, polygon, and even dominant\nclass labeling in terms of annotation time per click. However, it introduces\nthe class ambiguity issue in training as it assigns partial labels (i.e., a set\nof candidate classes) to individual pixels. We thus propose a new algorithm for\nlearning semantic segmentation while disambiguating the partial labels in two\nstages. In the first stage, it trains a segmentation model directly with the\npartial labels through two new loss functions motivated by partial label\nlearning and multiple instance learning. In the second stage, it disambiguates\nthe partial labels by generating pixel-wise pseudo labels, which are used for\nsupervised learning of the model. Equipped with a new acquisition function\ndedicated to the multi-class labeling, our method outperforms previous work on\nCityscapes and PASCAL VOC 2012 while spending less annotation cost. Our code\nand results are available at https://github.com/sehyun03/MulActSeg.\n", "date": "2023-09-17", "categories": []}
{"id": "2311.02345", "title": "Perturbation-based Active Learning for Question Answering", "abstract": "  Building a question answering (QA) model with less annotation costs can be\nachieved by utilizing active learning (AL) training strategy. It selects the\nmost informative unlabeled training data to update the model effectively.\nAcquisition functions for AL are used to determine how informative each\ntraining example is, such as uncertainty or diversity based sampling. In this\nwork, we propose a perturbation-based active learning acquisition strategy and\ndemonstrate it is more effective than existing commonly used strategies.\n", "date": "2023-11-04", "categories": []}
{"id": "2301.10625", "title": "Navigating the Pitfalls of Active Learning Evaluation: A Systematic\n  Framework for Meaningful Performance Assessment", "abstract": "  Active Learning (AL) aims to reduce the labeling burden by interactively\nselecting the most informative samples from a pool of unlabeled data. While\nthere has been extensive research on improving AL query methods in recent\nyears, some studies have questioned the effectiveness of AL compared to\nemerging paradigms such as semi-supervised (Semi-SL) and self-supervised\nlearning (Self-SL), or a simple optimization of classifier configurations.\nThus, today's AL literature presents an inconsistent and contradictory\nlandscape, leaving practitioners uncertain about whether and how to use AL in\ntheir tasks. In this work, we make the case that this inconsistency arises from\na lack of systematic and realistic evaluation of AL methods. Specifically, we\nidentify five key pitfalls in the current literature that reflect the delicate\nconsiderations required for AL evaluation. Further, we present an evaluation\nframework that overcomes these pitfalls and thus enables meaningful statements\nabout the performance of AL methods. To demonstrate the relevance of our\nprotocol, we present a large-scale empirical study and benchmark for image\nclassification spanning various data sets, query methods, AL settings, and\ntraining paradigms. Our findings clarify the inconsistent picture in the\nliterature and enable us to give hands-on recommendations for practitioners.\nThe benchmark is hosted at https://github.com/IML-DKFZ/realistic-al .\n", "date": "2023-01-25", "categories": []}
{"id": "2311.02061", "title": "Active Learning-Based Species Range Estimation", "abstract": "  We propose a new active learning approach for efficiently estimating the\ngeographic range of a species from a limited number of on the ground\nobservations. We model the range of an unmapped species of interest as the\nweighted combination of estimated ranges obtained from a set of different\nspecies. We show that it is possible to generate this candidate set of ranges\nby using models that have been trained on large weakly supervised community\ncollected observation data. From this, we develop a new active querying\napproach that sequentially selects geographic locations to visit that best\nreduce our uncertainty over an unmapped species' range. We conduct a detailed\nevaluation of our approach and compare it to existing active learning methods\nusing an evaluation dataset containing expert-derived ranges for one thousand\nspecies. Our results demonstrate that our method outperforms alternative active\nlearning methods and approaches the performance of end-to-end trained models,\neven when only using a fraction of the data. This highlights the utility of\nactive learning via transfer learned spatial representations for species range\nestimation. It also emphasizes the value of leveraging emerging large-scale\ncrowdsourced datasets, not only for modeling a species' range, but also for\nactively discovering them.\n", "date": "2023-11-03", "categories": []}
{"id": "2311.00906", "title": "Re-weighting Tokens: A Simple and Effective Active Learning Strategy for\n  Named Entity Recognition", "abstract": "  Active learning, a widely adopted technique for enhancing machine learning\nmodels in text and image classification tasks with limited annotation\nresources, has received relatively little attention in the domain of Named\nEntity Recognition (NER). The challenge of data imbalance in NER has hindered\nthe effectiveness of active learning, as sequence labellers lack sufficient\nlearning signals. To address these challenges, this paper presents a novel\nreweighting-based active learning strategy that assigns dynamic smoothed\nweights to individual tokens. This adaptable strategy is compatible with\nvarious token-level acquisition functions and contributes to the development of\nrobust active learners. Experimental results on multiple corpora demonstrate\nthe substantial performance improvement achieved by incorporating our\nre-weighting strategy into existing acquisition functions, validating its\npractical efficacy.\n", "date": "2023-11-02", "categories": []}
{"id": "2310.19596", "title": "LLMaAA: Making Large Language Models as Active Annotators", "abstract": "  Prevalent supervised learning methods in natural language processing (NLP)\nare notoriously data-hungry, which demand large amounts of high-quality\nannotated data. In practice, acquiring such data is a costly endeavor.\nRecently, the superior few-shot performance of large language models (LLMs) has\npropelled the development of dataset generation, where the training data are\nsolely synthesized from LLMs. However, such an approach usually suffers from\nlow-quality issues, and requires orders of magnitude more labeled data to\nachieve satisfactory performance. To fully exploit the potential of LLMs and\nmake use of massive unlabeled data, we propose LLMaAA, which takes LLMs as\nannotators and puts them into an active learning loop to determine what to\nannotate efficiently. To learn robustly with pseudo labels, we optimize both\nthe annotation and training processes: (1) we draw k-NN examples from a small\ndemonstration pool as in-context examples, and (2) we adopt the example\nreweighting technique to assign training samples with learnable weights.\nCompared with previous approaches, LLMaAA features both efficiency and\nreliability. We conduct experiments and analysis on two classic NLP tasks,\nnamed entity recognition and relation extraction. With LLMaAA, task-specific\nmodels trained from LLM-generated labels can outperform the teacher within only\nhundreds of annotated examples, which is much more cost-effective than other\nbaselines.\n", "date": "2023-10-30", "categories": []}
{"id": "2310.20153", "title": "Interactive Multi-fidelity Learning for Cost-effective Adaptation of\n  Language Model with Sparse Human Supervision", "abstract": "  Large language models (LLMs) have demonstrated remarkable capabilities in\nvarious tasks. However, their suitability for domain-specific tasks, is limited\ndue to their immense scale at deployment, susceptibility to misinformation, and\nmore importantly, high data annotation costs. We propose a novel Interactive\nMulti-Fidelity Learning (IMFL) framework for the cost-effective development of\nsmall domain-specific LMs under limited annotation budgets. Our approach\nformulates the domain-specific fine-tuning process as a multi-fidelity learning\nproblem, focusing on identifying the optimal acquisition strategy that balances\nbetween low-fidelity automatic LLM annotations and high-fidelity human\nannotations to maximize model performance. We further propose an\nexploration-exploitation query strategy that enhances annotation diversity and\ninformativeness, incorporating two innovative designs: 1) prompt retrieval that\nselects in-context examples from human-annotated samples to improve LLM\nannotation, and 2) variable batch size that controls the order for choosing\neach fidelity to facilitate knowledge distillation, ultimately enhancing\nannotation quality. Extensive experiments on financial and medical tasks\ndemonstrate that IMFL achieves superior performance compared with single\nfidelity annotations. Given a limited budget of human annotation, IMFL\nsignificantly outperforms the human annotation baselines in all four tasks and\nachieves very close performance as human annotations on two of the tasks. These\npromising results suggest that the high human annotation costs in\ndomain-specific tasks can be significantly reduced by employing IMFL, which\nutilizes fewer human annotations, supplemented with cheaper and faster LLM\n(e.g., GPT-3.5) annotations to achieve comparable performance.\n", "date": "2023-10-31", "categories": []}
{"id": "2310.20293", "title": "Annotator: A Generic Active Learning Baseline for LiDAR Semantic\n  Segmentation", "abstract": "  Active learning, a label-efficient paradigm, empowers models to interactively\nquery an oracle for labeling new data. In the realm of LiDAR semantic\nsegmentation, the challenges stem from the sheer volume of point clouds,\nrendering annotation labor-intensive and cost-prohibitive. This paper presents\nAnnotator, a general and efficient active learning baseline, in which a\nvoxel-centric online selection strategy is tailored to efficiently probe and\nannotate the salient and exemplar voxel girds within each LiDAR scan, even\nunder distribution shift. Concretely, we first execute an in-depth analysis of\nseveral common selection strategies such as Random, Entropy, Margin, and then\ndevelop voxel confusion degree (VCD) to exploit the local topology relations\nand structures of point clouds. Annotator excels in diverse settings, with a\nparticular focus on active learning (AL), active source-free domain adaptation\n(ASFDA), and active domain adaptation (ADA). It consistently delivers\nexceptional performance across LiDAR semantic segmentation benchmarks, spanning\nboth simulation-to-real and real-to-real scenarios. Surprisingly, Annotator\nexhibits remarkable efficiency, requiring significantly fewer annotations,\ne.g., just labeling five voxels per scan in the SynLiDAR-to-SemanticKITTI task.\nThis results in impressive performance, achieving 87.8% fully-supervised\nperformance under AL, 88.5% under ASFDA, and 94.4% under ADA. We envision that\nAnnotator will offer a simple, general, and efficient solution for\nlabel-efficient 3D applications. Project page:\nhttps://binhuixie.github.io/annotator-web\n", "date": "2023-10-31", "categories": []}
{"id": "2211.15788", "title": "A Visual Active Search Framework for Geospatial Exploration", "abstract": "  Many problems can be viewed as forms of geospatial search aided by aerial\nimagery, with examples ranging from detecting poaching activity to human\ntrafficking. We model this class of problems in a visual active search (VAS)\nframework, which has three key inputs: (1) an image of the entire search area,\nwhich is subdivided into regions, (2) a local search function, which determines\nwhether a previously unseen object class is present in a given region, and (3)\na fixed search budget, which limits the number of times the local search\nfunction can be evaluated. The goal is to maximize the number of objects found\nwithin the search budget. We propose a reinforcement learning approach for VAS\nthat learns a meta-search policy from a collection of fully annotated search\ntasks. This meta-search policy is then used to dynamically search for a novel\ntarget-object class, leveraging the outcome of any previous queries to\ndetermine where to query next. Through extensive experiments on several\nlarge-scale satellite imagery datasets, we show that the proposed approach\nsignificantly outperforms several strong baselines. We also propose novel\ndomain adaptation techniques that improve the policy at decision time when\nthere is a significant domain gap with the training data. Code is publicly\navailable.\n", "date": "2022-11-28", "categories": []}
{"id": "2306.12371", "title": "Optimistic Active Exploration of Dynamical Systems", "abstract": "  Reinforcement learning algorithms commonly seek to optimize policies for\nsolving one particular task. How should we explore an unknown dynamical system\nsuch that the estimated model globally approximates the dynamics and allows us\nto solve multiple downstream tasks in a zero-shot manner? In this paper, we\naddress this challenge, by developing an algorithm -- OPAX -- for active\nexploration. OPAX uses well-calibrated probabilistic models to quantify the\nepistemic uncertainty about the unknown dynamics. It optimistically -- w.r.t.\nto plausible dynamics -- maximizes the information gain between the unknown\ndynamics and state observations. We show how the resulting optimization problem\ncan be reduced to an optimal control problem that can be solved at each episode\nusing standard approaches. We analyze our algorithm for general models, and, in\nthe case of Gaussian process dynamics, we give a first-of-its-kind sample\ncomplexity bound and show that the epistemic uncertainty converges to zero. In\nour experiments, we compare OPAX with other heuristic active exploration\napproaches on several environments. Our experiments show that OPAX is not only\ntheoretically sound but also performs well for zero-shot planning on novel\ndownstream tasks.\n", "date": "2023-06-21", "categories": []}
{"id": "2307.03899", "title": "Active Learning in Physics: From 101, to Progress, and Perspective", "abstract": "  Active Learning (AL) is a family of machine learning (ML) algorithms that\npredates the current era of artificial intelligence. Unlike traditional\napproaches that require labeled samples for training, AL iteratively selects\nunlabeled samples to be annotated by an expert. This protocol aims to\nprioritize the most informative samples, leading to improved model performance\ncompared to training with all labeled samples. In recent years, AL has gained\nincreasing attention, particularly in the field of physics. This paper presents\na comprehensive and accessible introduction to the theory of AL reviewing the\nlatest advancements across various domains. Additionally, we explore the\npotential integration of AL with quantum ML, envisioning a synergistic fusion\nof these two fields rather than viewing AL as a mere extension of classical ML\ninto the quantum realm.\n", "date": "2023-07-08", "categories": []}
{"id": "2310.10893", "title": "Active Learning Framework for Cost-Effective TCR-Epitope Binding\n  Affinity Prediction", "abstract": "  T cell receptors (TCRs) are critical components of adaptive immune systems,\nresponsible for responding to threats by recognizing epitope sequences\npresented on host cell surface. Computational prediction of binding affinity\nbetween TCRs and epitope sequences using machine/deep learning has attracted\nintense attention recently. However, its success is hindered by the lack of\nlarge collections of annotated TCR-epitope pairs. Annotating their binding\naffinity requires expensive and time-consuming wet-lab evaluation. To reduce\nannotation cost, we present ActiveTCR, a framework that incorporates active\nlearning and TCR-epitope binding affinity prediction models. Starting with a\nsmall set of labeled training pairs, ActiveTCR iteratively searches for\nunlabeled TCR-epitope pairs that are ''worth'' for annotation. It aims to\nmaximize performance gains while minimizing the cost of annotation. We compared\nfour query strategies with a random sampling baseline and demonstrated that\nActiveTCR reduces annotation costs by approximately 40%. Furthermore, we showed\nthat providing ground truth labels of TCR-epitope pairs to query strategies can\nhelp identify and reduce more than 40% redundancy among already annotated pairs\nwithout compromising model performance, enabling users to train equally\npowerful prediction models with less training data. Our work is the first\nsystematic investigation of data optimization for TCR-epitope binding affinity\nprediction.\n", "date": "2023-10-16", "categories": []}
{"id": "2310.19573", "title": "Model Uncertainty based Active Learning on Tabular Data using Boosted\n  Trees", "abstract": "  Supervised machine learning relies on the availability of good labelled data\nfor model training. Labelled data is acquired by human annotation, which is a\ncumbersome and costly process, often requiring subject matter experts. Active\nlearning is a sub-field of machine learning which helps in obtaining the\nlabelled data efficiently by selecting the most valuable data instances for\nmodel training and querying the labels only for those instances from the human\nannotator. Recently, a lot of research has been done in the field of active\nlearning, especially for deep neural network based models. Although deep\nlearning shines when dealing with image\\textual\\multimodal data, gradient\nboosting methods still tend to achieve much better results on tabular data. In\nthis work, we explore active learning for tabular data using boosted trees.\nUncertainty based sampling in active learning is the most commonly used\nquerying strategy, wherein the labels of those instances are sequentially\nqueried for which the current model prediction is maximally uncertain. Entropy\nis often the choice for measuring uncertainty. However, entropy is not exactly\na measure of model uncertainty. Although there has been a lot of work in deep\nlearning for measuring model uncertainty and employing it in active learning,\nit is yet to be explored for non-neural network models. To this end, we explore\nthe effectiveness of boosted trees based model uncertainty methods in active\nlearning. Leveraging this model uncertainty, we propose an uncertainty based\nsampling in active learning for regression tasks on tabular data. Additionally,\nwe also propose a novel cost-effective active learning method for regression\ntasks along with an improved cost-effective active learning method for\nclassification tasks.\n", "date": "2023-10-30", "categories": []}
{"id": "2310.17044", "title": "Learning to Rank for Active Learning via Multi-Task Bilevel Optimization", "abstract": "  Active learning is a promising paradigm to reduce the labeling cost by\nstrategically requesting labels to improve model performance. However, existing\nactive learning methods often rely on expensive acquisition function to\ncompute, extensive modeling retraining and multiple rounds of interaction with\nannotators. To address these limitations, we propose a novel approach for\nactive learning, which aims to select batches of unlabeled instances through a\nlearned surrogate model for data acquisition. A key challenge in this approach\nis developing an acquisition function that generalizes well, as the history of\ndata, which forms part of the utility function's input, grows over time. Our\nnovel algorithmic contribution is a bilevel multi-task bilevel optimization\nframework that predicts the relative utility -- measured by the validation\naccuracy -- of different training sets, and ensures the learned acquisition\nfunction generalizes effectively. For cases where validation accuracy is\nexpensive to evaluate, we introduce efficient interpolation-based surrogate\nmodels to estimate the utility function, reducing the evaluation cost. We\ndemonstrate the performance of our approach through extensive experiments on\nstandard active classification benchmarks. By employing our learned utility\nfunction, we show significant improvements over traditional techniques, paving\nthe way for more efficient and effective utility maximization in active\nlearning applications.\n", "date": "2023-10-25", "categories": []}
{"id": "2212.11680", "title": "Smooth Sailing: Improving Active Learning for Pre-trained Language\n  Models with Representation Smoothness Analysis", "abstract": "  Developed to alleviate prohibitive labeling costs, active learning (AL)\nmethods aim to reduce label complexity in supervised learning. While recent\nwork has demonstrated the benefit of using AL in combination with large\npre-trained language models (PLMs), it has often overlooked the practical\nchallenges that hinder the effectiveness of AL. We address these challenges by\nleveraging representation smoothness analysis to ensure AL is feasible, that\nis, both effective and practicable. Firstly, we propose an early stopping\ntechnique that does not require a validation set -- often unavailable in\nrealistic AL conditions -- and observe significant improvements over random\nsampling across multiple datasets and AL methods. Further, we find that task\nadaptation improves AL, whereas standard short fine-tuning in AL does not\nprovide improvements over random sampling. Our work demonstrates the usefulness\nof representation smoothness analysis for AL and introduces an AL stopping\ncriterion that reduces label complexity.\n", "date": "2022-12-20", "categories": []}
{"id": "2305.12710", "title": "Beyond Labels: Empowering Human Annotators with Natural Language\n  Explanations through a Novel Active-Learning Architecture", "abstract": "  Real-world domain experts (e.g., doctors) rarely annotate only a decision\nlabel in their day-to-day workflow without providing explanations. Yet,\nexisting low-resource learning techniques, such as Active Learning (AL), that\naim to support human annotators mostly focus on the label while neglecting the\nnatural language explanation of a data point. This work proposes a novel AL\narchitecture to support experts' real-world need for label and explanation\nannotations in low-resource scenarios. Our AL architecture leverages an\nexplanation-generation model to produce explanations guided by human\nexplanations, a prediction model that utilizes generated explanations toward\nprediction faithfully, and a novel data diversity-based AL sampling strategy\nthat benefits from the explanation annotations. Automated and human evaluations\ndemonstrate the effectiveness of incorporating explanations into AL sampling\nand the improved human annotation efficiency and trustworthiness with our AL\narchitecture. Additional ablation studies illustrate the potential of our AL\narchitecture for transfer learning, generalizability, and integration with\nlarge language models (LLMs). While LLMs exhibit exceptional\nexplanation-generation capabilities for relatively simple tasks, their\neffectiveness in complex real-world tasks warrants further in-depth study.\n", "date": "2023-05-22", "categories": []}
{"id": "2310.14513", "title": "Turn-Level Active Learning for Dialogue State Tracking", "abstract": "  Dialogue state tracking (DST) plays an important role in task-oriented\ndialogue systems. However, collecting a large amount of turn-by-turn annotated\ndialogue data is costly and inefficient. In this paper, we propose a novel\nturn-level active learning framework for DST to actively select turns in\ndialogues to annotate. Given the limited labelling budget, experimental results\ndemonstrate the effectiveness of selective annotation of dialogue turns.\nAdditionally, our approach can effectively achieve comparable DST performance\nto traditional training approaches with significantly less annotated data,\nwhich provides a more efficient way to annotate new dialogue data.\n", "date": "2023-10-23", "categories": []}
{"id": "2310.14979", "title": "ACTOR: Active Learning with Annotator-specific Classification Heads to\n  Embrace Human Label Variation", "abstract": "  Label aggregation such as majority voting is commonly used to resolve\nannotator disagreement in dataset creation. However, this may disregard\nminority values and opinions. Recent studies indicate that learning from\nindividual annotations outperforms learning from aggregated labels, though they\nrequire a considerable amount of annotation. Active learning, as an annotation\ncost-saving strategy, has not been fully explored in the context of learning\nfrom disagreement. We show that in the active learning setting, a multi-head\nmodel performs significantly better than a single-head model in terms of\nuncertainty estimation. By designing and evaluating acquisition functions with\nannotator-specific heads on two datasets, we show that group-level entropy\nworks generally well on both datasets. Importantly, it achieves performance in\nterms of both prediction and uncertainty estimation comparable to full-scale\ntraining from disagreement, while saving up to 70% of the annotation budget.\n", "date": "2023-10-23", "categories": []}
{"id": "2305.12634", "title": "Data-efficient Active Learning for Structured Prediction with Partial\n  Annotation and Self-Training", "abstract": "  In this work we propose a pragmatic method that reduces the annotation cost\nfor structured label spaces using active learning. Our approach leverages\npartial annotation, which reduces labeling costs for structured outputs by\nselecting only the most informative sub-structures for annotation. We also\nutilize self-training to incorporate the current model's automatic predictions\nas pseudo-labels for un-annotated sub-structures. A key challenge in\neffectively combining partial annotation with self-training to reduce\nannotation cost is determining which sub-structures to select to label. To\naddress this challenge, we adopt an error estimator to adaptively decide the\npartial selection ratio according to the current model's capability. In\nevaluations spanning four structured prediction tasks, we show that our\ncombination of partial annotation and self-training using an adaptive selection\nratio reduces annotation cost over strong full annotation baselines under a\nfair comparison scheme that takes reading time into consideration.\n", "date": "2023-05-22", "categories": []}
{"id": "2305.15040", "title": "Active Learning for Natural Language Generation", "abstract": "  The field of Natural Language Generation (NLG) suffers from a severe shortage\nof labeled data due to the extremely expensive and time-consuming process\ninvolved in manual annotation. A natural approach for coping with this problem\nis active learning (AL), a well-known machine learning technique for improving\nannotation efficiency by selectively choosing the most informative examples to\nlabel. However, while AL has been well-researched in the context of text\nclassification, its application to NLG remains largely unexplored. In this\npaper, we present a first systematic study of active learning for NLG,\nconsidering a diverse set of tasks and multiple leading selection strategies,\nand harnessing a strong instruction-tuned model. Our results indicate that the\nperformance of existing AL strategies is inconsistent, surpassing the baseline\nof random example selection in some cases but not in others. We highlight some\nnotable differences between the classification and generation scenarios, and\nanalyze the selection behaviors of existing AL strategies. Our findings\nmotivate exploring novel approaches for applying AL to generation tasks.\n", "date": "2023-05-24", "categories": []}
{"id": "2309.17342", "title": "Towards Free Data Selection with General-Purpose Models", "abstract": "  A desirable data selection algorithm can efficiently choose the most\ninformative samples to maximize the utility of limited annotation budgets.\nHowever, current approaches, represented by active learning methods, typically\nfollow a cumbersome pipeline that iterates the time-consuming model training\nand batch data selection repeatedly. In this paper, we challenge this status\nquo by designing a distinct data selection pipeline that utilizes existing\ngeneral-purpose models to select data from various datasets with a single-pass\ninference without the need for additional training or supervision. A novel free\ndata selection (FreeSel) method is proposed following this new pipeline.\nSpecifically, we define semantic patterns extracted from inter-mediate features\nof the general-purpose model to capture subtle local information in each image.\nWe then enable the selection of all data samples in a single pass through\ndistance-based sampling at the fine-grained semantic pattern level. FreeSel\nbypasses the heavy batch selection process, achieving a significant improvement\nin efficiency and being 530x faster than existing active learning methods.\nExtensive experiments verify the effectiveness of FreeSel on various computer\nvision tasks. Our code is available at https://github.com/yichen928/FreeSel.\n", "date": "2023-09-29", "categories": []}
{"id": "2310.06306", "title": "Ensemble Active Learning by Contextual Bandits for AI Incubation in\n  Manufacturing", "abstract": "  It is challenging but important to save annotation efforts in streaming data\nacquisition to maintain data quality for supervised learning base learners. We\npropose an ensemble active learning method to actively acquire samples for\nannotation by contextual bandits, which is will enforce the\nexploration-exploitation balance and leading to improved AI modeling\nperformance.\n", "date": "2023-10-10", "categories": []}
{"id": "2310.04461", "title": "AI-based automated active learning for discovery of hidden dynamic\n  processes: A use case in light microscopy", "abstract": "  In the biomedical environment, experiments assessing dynamic processes are\nprimarily performed by a human acquisition supervisor. Contemporary\nimplementations of such experiments frequently aim to acquire a maximum number\nof relevant events from sometimes several hundred parallel, non-synchronous\nprocesses. Since in some high-throughput experiments, only one or a few\ninstances of a given process can be observed simultaneously, a strategy for\nplanning and executing an efficient acquisition paradigm is essential. To\naddress this problem, we present two new methods in this paper. The first\nmethod, Encoded Dynamic Process (EDP), is Artificial Intelligence (AI)-based\nand represents dynamic processes so as to allow prediction of pseudo-time\nvalues from single still images. Second, with Experiment Automation Pipeline\nfor Dynamic Processes (EAPDP), we present a Machine Learning Operations\n(MLOps)-based pipeline that uses the extracted knowledge from EDP to\nefficiently schedule acquisition in biomedical experiments for dynamic\nprocesses in practice. In a first experiment, we show that the pre-trained\nState-Of-The- Art (SOTA) object segmentation method Contour Proposal Networks\n(CPN) works reliably as a module of EAPDP to extract the relevant object for\nEDP from the acquired three-dimensional image stack.\n", "date": "2023-10-05", "categories": []}
{"id": "2310.05018", "title": "Human-in-the-loop: The future of Machine Learning in Automated Electron\n  Microscopy", "abstract": "  Machine learning methods are progressively gaining acceptance in the electron\nmicroscopy community for de-noising, semantic segmentation, and dimensionality\nreduction of data post-acquisition. The introduction of the APIs by major\ninstrument manufacturers now allows the deployment of ML workflows in\nmicroscopes, not only for data analytics but also for real-time decision-making\nand feedback for microscope operation. However, the number of use cases for\nreal-time ML remains remarkably small. Here, we discuss some considerations in\ndesigning ML-based active experiments and pose that the likely strategy for the\nnext several years will be human-in-the-loop automated experiments (hAE). In\nthis paradigm, the ML learning agent directly controls beam position and image\nand spectroscopy acquisition functions, and human operator monitors experiment\nprogression in real- and feature space of the system and tunes the policies of\nthe ML agent to steer the experiment towards specific objectives.\n", "date": "2023-10-08", "categories": []}
{"id": "2310.01597", "title": "Pool-Based Active Learning with Proper Topological Regions", "abstract": "  Machine learning methods usually rely on large sample size to have good\nperformance, while it is difficult to provide labeled set in many applications.\nPool-based active learning methods are there to detect, among a set of\nunlabeled data, the ones that are the most relevant for the training. We\npropose in this paper a meta-approach for pool-based active learning strategies\nin the context of multi-class classification tasks based on Proper Topological\nRegions. PTR, based on topological data analysis (TDA), are relevant regions\nused to sample cold-start points or within the active learning scheme. The\nproposed method is illustrated empirically on various benchmark datasets, being\ncompetitive to the classical methods from the literature.\n", "date": "2023-10-02", "categories": []}
{"id": "2211.13579", "title": "Knowledge-Aware Federated Active Learning with Non-IID Data", "abstract": "  Federated learning enables multiple decentralized clients to learn\ncollaboratively without sharing the local training data. However, the expensive\nannotation cost to acquire data labels on local clients remains an obstacle in\nutilizing local data. In this paper, we propose a federated active learning\nparadigm to efficiently learn a global model with limited annotation budget\nwhile protecting data privacy in a decentralized learning way. The main\nchallenge faced by federated active learning is the mismatch between the active\nsampling goal of the global model on the server and that of the asynchronous\nlocal clients. This becomes even more significant when data is distributed\nnon-IID across local clients. To address the aforementioned challenge, we\npropose Knowledge-Aware Federated Active Learning (KAFAL), which consists of\nKnowledge-Specialized Active Sampling (KSAS) and Knowledge-Compensatory\nFederated Update (KCFU). KSAS is a novel active sampling method tailored for\nthe federated active learning problem. It deals with the mismatch challenge by\nsampling actively based on the discrepancies between local and global models.\nKSAS intensifies specialized knowledge in local clients, ensuring the sampled\ndata to be informative for both the local clients and the global model. KCFU,\nin the meantime, deals with the client heterogeneity caused by limited data and\nnon-IID data distributions. It compensates for each client's ability in weak\nclasses by the assistance of the global model. Extensive experiments and\nanalyses are conducted to show the superiority of KSAS over the\nstate-of-the-art active learning methods and the efficiency of KCFU under the\nfederated active learning framework.\n", "date": "2022-11-24", "categories": []}
{"id": "2304.07665", "title": "Dynamic Exploration-Exploitation Trade-Off in Active Learning Regression\n  with Bayesian Hierarchical Modeling", "abstract": "  Active learning provides a framework to adaptively query the most informative\nexperiments towards learning an unknown black-box function. Various approaches\nof active learning have been proposed in the literature, however, they either\nfocus on exploration or exploitation in the design space. Methods that do\nconsider exploration-exploitation simultaneously employ fixed or ad-hoc\nmeasures to control the trade-off that may not be optimal. In this paper, we\ndevelop a Bayesian hierarchical approach, referred as BHEEM, to dynamically\nbalance the exploration-exploitation trade-off as more data points are queried.\nTo sample from the posterior distribution of the trade-off parameter, We\nsubsequently formulate an approximate Bayesian computation approach based on\nthe linear dependence of queried data in the feature space. Simulated and\nreal-world examples show the proposed approach achieves at least 21% and 11%\naverage improvement when compared to pure exploration and exploitation\nstrategies respectively. More importantly, we note that by optimally balancing\nthe trade-off between exploration and exploitation, BHEEM performs better or at\nleast as well as either pure exploration or pure exploitation.\n", "date": "2023-04-16", "categories": []}
{"id": "2308.00146", "title": "DiffusAL: Coupling Active Learning with Graph Diffusion for\n  Label-Efficient Node Classification", "abstract": "  Node classification is one of the core tasks on attributed graphs, but\nsuccessful graph learning solutions require sufficiently labeled data. To keep\nannotation costs low, active graph learning focuses on selecting the most\nqualitative subset of nodes that maximizes label efficiency. However, deciding\nwhich heuristic is best suited for an unlabeled graph to increase label\nefficiency is a persistent challenge. Existing solutions either neglect\naligning the learned model and the sampling method or focus only on limited\nselection aspects. They are thus sometimes worse or only equally good as random\nsampling. In this work, we introduce a novel active graph learning approach\ncalled DiffusAL, showing significant robustness in diverse settings. Toward\nbetter transferability between different graph structures, we combine three\nindependent scoring functions to identify the most informative node samples for\nlabeling in a parameter-free way: i) Model Uncertainty, ii) Diversity\nComponent, and iii) Node Importance computed via graph diffusion heuristics.\nMost of our calculations for acquisition and training can be pre-processed,\nmaking DiffusAL more efficient compared to approaches combining diverse\nselection criteria and similarly fast as simpler heuristics. Our experiments on\nvarious benchmark datasets show that, unlike previous methods, our approach\nsignificantly outperforms random selection in 100% of all datasets and labeling\nbudgets tested.\n", "date": "2023-07-31", "categories": []}
{"id": "2310.00372", "title": "Deep Active Learning with Noisy Oracle in Object Detection", "abstract": "  Obtaining annotations for complex computer vision tasks such as object\ndetection is an expensive and time-intense endeavor involving a large number of\nhuman workers or expert opinions. Reducing the amount of annotations required\nwhile maintaining algorithm performance is, therefore, desirable for machine\nlearning practitioners and has been successfully achieved by active learning\nalgorithms. However, it is not merely the amount of annotations which\ninfluences model performance but also the annotation quality. In practice, the\noracles that are queried for new annotations frequently contain significant\namounts of noise. Therefore, cleansing procedures are oftentimes necessary to\nreview and correct given labels. This process is subject to the same budget as\nthe initial annotation itself since it requires human workers or even domain\nexperts. Here, we propose a composite active learning framework including a\nlabel review module for deep object detection. We show that utilizing part of\nthe annotation budget to correct the noisy annotations partially in the active\ndataset leads to early improvements in model performance, especially when\ncoupled with uncertainty-based query strategies. The precision of the label\nerror proposals has a significant influence on the measured effect of the label\nreview. In our experiments we achieve improvements of up to 4.5 mAP points of\nobject detection performance by incorporating label reviews at equal annotation\nbudget.\n", "date": "2023-09-30", "categories": []}
{"id": "2310.00511", "title": "Nonparametric active learning for cost-sensitive classification", "abstract": "  Cost-sensitive learning is a common type of machine learning problem where\ndifferent errors of prediction incur different costs. In this paper, we design\na generic nonparametric active learning algorithm for cost-sensitive\nclassification. Based on the construction of confidence bounds for the expected\nprediction cost functions of each label, our algorithm sequentially selects the\nmost informative vector points. Then it interacts with them by only querying\nthe costs of prediction that could be the smallest. We prove that our algorithm\nattains optimal rate of convergence in terms of the number of interactions with\nthe feature vector space. Furthermore, in terms of a general version of\nTsybakov's noise assumption, the gain over the corresponding passive learning\nis explicitly characterized by the probability-mass of the boundary decision.\nAdditionally, we prove the near-optimality of obtained upper bounds by\nproviding matching (up to logarithmic factor) lower bounds.\n", "date": "2023-09-30", "categories": []}
{"id": "2310.01003", "title": "Conflict-Aware Active Automata Learning", "abstract": "  Active automata learning algorithms cannot easily handle conflict in the\nobservation data (different outputs observed for the same inputs). This\ninherent inability to recover after a conflict impairs their effective\napplicability in scenarios where noise is present or the system under learning\nis mutating. We propose the Conflict-Aware Active Automata Learning (C3AL)\nframework to enable handling conflicting information during the learning\nprocess. The core idea is to consider the so-called observation tree as a\nfirst-class citizen in the learning process. Though this idea is explored in\nrecent work, we take it to its full effect by enabling its use with any\nexisting learner and minimizing the number of tests performed on the system\nunder learning, specially in the face of conflicts. We evaluate C3AL in a large\nset of benchmarks, covering over 30 different realistic targets, and over\n18,000 different scenarios. The results of the evaluation show that C3AL is a\nsuitable alternative framework for closed-box learning that can better handle\nnoise and mutations.\n", "date": "2023-10-02", "categories": []}
{"id": "2210.06788", "title": "TiDAL: Learning Training Dynamics for Active Learning", "abstract": "  Active learning (AL) aims to select the most useful data samples from an\nunlabeled data pool and annotate them to expand the labeled dataset under a\nlimited budget. Especially, uncertainty-based methods choose the most uncertain\nsamples, which are known to be effective in improving model performance.\nHowever, AL literature often overlooks training dynamics (TD), defined as the\never-changing model behavior during optimization via stochastic gradient\ndescent, even though other areas of literature have empirically shown that TD\nprovides important clues for measuring the sample uncertainty. In this paper,\nwe propose a novel AL method, Training Dynamics for Active Learning (TiDAL),\nwhich leverages the TD to quantify uncertainties of unlabeled data. Since\ntracking the TD of all the large-scale unlabeled data is impractical, TiDAL\nutilizes an additional prediction module that learns the TD of labeled data. To\nfurther justify the design of TiDAL, we provide theoretical and empirical\nevidence to argue the usefulness of leveraging TD for AL. Experimental results\nshow that our TiDAL achieves better or comparable performance on both balanced\nand imbalanced benchmark datasets compared to state-of-the-art AL methods,\nwhich estimate data uncertainty using only static information after model\ntraining.\n", "date": "2022-10-13", "categories": []}
{"id": "2303.15256", "title": "Active Self-Supervised Learning: A Few Low-Cost Relationships Are All\n  You Need", "abstract": "  Self-Supervised Learning (SSL) has emerged as the solution of choice to learn\ntransferable representations from unlabeled data. However, SSL requires to\nbuild samples that are known to be semantically akin, i.e. positive views.\nRequiring such knowledge is the main limitation of SSL and is often tackled by\nad-hoc strategies e.g. applying known data-augmentations to the same input. In\nthis work, we formalize and generalize this principle through Positive Active\nLearning (PAL) where an oracle queries semantic relationships between samples.\nPAL achieves three main objectives. First, it unveils a theoretically grounded\nlearning framework beyond SSL, based on similarity graphs, that can be extended\nto tackle supervised and semi-supervised learning depending on the employed\noracle. Second, it provides a consistent algorithm to embed a priori knowledge,\ne.g. some observed labels, into any SSL losses without any change in the\ntraining pipeline. Third, it provides a proper active learning framework\nyielding low-cost solutions to annotate datasets, arguably bringing the gap\nbetween theory and practice of active learning that is based on\nsimple-to-answer-by-non-experts queries of semantic relationships between\ninputs.\n", "date": "2023-03-27", "categories": []}
{"id": "2305.09807", "title": "On Dataset Transferability in Active Learning for Transformers", "abstract": "  Active learning (AL) aims to reduce labeling costs by querying the examples\nmost beneficial for model learning. While the effectiveness of AL for\nfine-tuning transformer-based pre-trained language models (PLMs) has been\ndemonstrated, it is less clear to what extent the AL gains obtained with one\nmodel transfer to others. We consider the problem of transferability of\nactively acquired datasets in text classification and investigate whether AL\ngains persist when a dataset built using AL coupled with a specific PLM is used\nto train a different PLM. We link the AL dataset transferability to the\nsimilarity of instances queried by the different PLMs and show that AL methods\nwith similar acquisition sequences produce highly transferable datasets\nregardless of the models used. Additionally, we show that the similarity of\nacquisition sequences is influenced more by the choice of the AL method than\nthe choice of the model.\n", "date": "2023-05-16", "categories": []}
{"id": "2309.14460", "title": "Online Active Learning For Sound Event Detection", "abstract": "  Data collection and annotation is a laborious, time-consuming prerequisite\nfor supervised machine learning tasks. Online Active Learning (OAL) is a\nparadigm that addresses this issue by simultaneously minimizing the amount of\nannotation required to train a classifier and adapting to changes in the data\nover the duration of the data collection process. Prior work has indicated that\nfluctuating class distributions and data drift are still common problems for\nOAL. This work presents new loss functions that address these challenges when\nOAL is applied to Sound Event Detection (SED). Experimental results from the\nSONYC dataset and two Voice-Type Discrimination (VTD) corpora indicate that OAL\ncan reduce the time and effort required to train SED classifiers by a factor of\n5 for SONYC, and that the new methods presented here successfully resolve\nissues present in existing OAL methods.\n", "date": "2023-09-25", "categories": []}
{"id": "2309.16139", "title": "Two-Step Active Learning for Instance Segmentation with Uncertainty and\n  Diversity Sampling", "abstract": "  Training high-quality instance segmentation models requires an abundance of\nlabeled images with instance masks and classifications, which is often\nexpensive to procure. Active learning addresses this challenge by striving for\noptimum performance with minimal labeling cost by selecting the most\ninformative and representative images for labeling. Despite its potential,\nactive learning has been less explored in instance segmentation compared to\nother tasks like image classification, which require less labeling. In this\nstudy, we propose a post-hoc active learning algorithm that integrates\nuncertainty-based sampling with diversity-based sampling. Our proposed\nalgorithm is not only simple and easy to implement, but it also delivers\nsuperior performance on various datasets. Its practical application is\ndemonstrated on a real-world overhead imagery dataset, where it increases the\nlabeling efficiency fivefold.\n", "date": "2023-09-28", "categories": []}
{"id": "2309.05517", "title": "Stream-based Active Learning by Exploiting Temporal Properties in\n  Perception with Temporal Predicted Loss", "abstract": "  Active learning (AL) reduces the amount of labeled data needed to train a\nmachine learning model by intelligently choosing which instances to label.\nClassic pool-based AL requires all data to be present in a datacenter, which\ncan be challenging with the increasing amounts of data needed in deep learning.\nHowever, AL on mobile devices and robots, like autonomous cars, can filter the\ndata from perception sensor streams before reaching the datacenter. We\nexploited the temporal properties for such image streams in our work and\nproposed the novel temporal predicted loss (TPL) method. To evaluate the\nstream-based setting properly, we introduced the GTA V streets and the A2D2\nstreets dataset and made both publicly available. Our experiments showed that\nour approach significantly improves the diversity of the selection while being\nan uncertainty-based method. As pool-based approaches are more common in\nperception applications, we derived a concept for comparing pool-based and\nstream-based AL, where TPL out-performed state-of-the-art pool- or stream-based\napproaches for different models. TPL demonstrated a gain of 2.5 precept points\n(pp) less required data while being significantly faster than pool-based\nmethods.\n", "date": "2023-09-11", "categories": []}
{"id": "2212.01165", "title": "Deep Active Learning for Multi-Label Classification of Remote Sensing\n  Images", "abstract": "  In this letter, we introduce deep active learning (AL) for multi-label\nclassification (MLC) problems in remote sensing (RS). In particular, we\ninvestigate the effectiveness of several AL query functions for MLC of RS\nimages. Unlike the existing AL query functions (which are defined for\nsingle-label classification or semantic segmentation problems), each query\nfunction in this paper is based on the evaluation of two criteria: i)\nmulti-label uncertainty; and ii) multi-label diversity. The multi-label\nuncertainty criterion is associated to the confidence of the deep neural\nnetworks (DNNs) in correctly assigning multi-labels to each image. To assess\nthis criterion, we investigate three strategies: i) learning multi-label loss\nordering; ii) measuring temporal discrepancy of multi-label predictions; and\niii) measuring magnitude of approximated gradient embeddings. The multi-label\ndiversity criterion is associated to the selection of a set of images that are\nas diverse as possible to each other that prevents redundancy among them. To\nassess this criterion, we exploit a clustering based strategy. We combine each\nof the above-mentioned uncertainty strategies with the clustering based\ndiversity strategy, resulting in three different query functions. All the\nconsidered query functions are introduced for the first time in the framework\nof MLC problems in RS. Experimental results obtained on two benchmark archives\nshow that these query functions result in the selection of a highly informative\nset of samples at each iteration of the AL process.\n", "date": "2022-12-02", "categories": []}
{"id": "2309.13276", "title": "Discwise Active Learning for LiDAR Semantic Segmentation", "abstract": "  While LiDAR data acquisition is easy, labeling for semantic segmentation\nremains highly time consuming and must therefore be done selectively. Active\nlearning (AL) provides a solution that can iteratively and intelligently label\na dataset while retaining high performance and a low budget. In this work we\nexplore AL for LiDAR semantic segmentation. As a human expert is a component of\nthe pipeline, a practical framework must consider common labeling techniques\nsuch as sequential labeling that drastically improve annotation times. We\ntherefore propose a discwise approach (DiAL), where in each iteration, we query\nthe region a single frame covers on global coordinates, labeling all frames\nsimultaneously. We then tackle the two major challenges that emerge with\ndiscwise AL. Firstly we devise a new acquisition function that takes 3D point\ndensity changes into consideration which arise due to location changes or\nego-vehicle motion. Next we solve a mixed-integer linear program that provides\na general solution to the selection of multiple frames while taking into\nconsideration the possibilities of disc intersections. Finally we propose a\nsemi-supervised learning approach to utilize all frames within our dataset and\nimprove performance.\n", "date": "2023-09-23", "categories": []}
{"id": "2212.02804", "title": "MUS-CDB: Mixed Uncertainty Sampling with Class Distribution Balancing\n  for Active Annotation in Aerial Object Detection", "abstract": "  Recent aerial object detection models rely on a large amount of labeled\ntraining data, which requires unaffordable manual labeling costs in large\naerial scenes with dense objects. Active learning effectively reduces the data\nlabeling cost by selectively querying the informative and representative\nunlabelled samples. However, existing active learning methods are mainly with\nclass-balanced settings and image-based querying for generic object detection\ntasks, which are less applicable to aerial object detection scenarios due to\nthe long-tailed class distribution and dense small objects in aerial scenes. In\nthis paper, we propose a novel active learning method for cost-effective aerial\nobject detection. Specifically, both object-level and image-level\ninformativeness are considered in the object selection to refrain from\nredundant and myopic querying. Besides, an easy-to-use class-balancing\ncriterion is incorporated to favor the minority objects to alleviate the\nlong-tailed class distribution problem in model training. We further devise a\ntraining loss to mine the latent knowledge in the unlabeled image regions.\nExtensive experiments are conducted on the DOTA-v1.0 and DOTA-v2.0 benchmarks\nto validate the effectiveness of the proposed method. For the ReDet, KLD, and\nSASM detectors on the DOTA-v2.0 dataset, the results show that our proposed\nMUS-CDB method can save nearly 75\\% of the labeling cost while achieving\ncomparable performance to other active learning methods in terms of mAP.Code is\npublicly online (https://github.com/ZJW700/MUS-CDB).\n", "date": "2022-12-06", "categories": []}
{"id": "2302.11075", "title": "Deep Active Learning in the Presence of Label Noise: A Survey", "abstract": "  Deep active learning has emerged as a powerful tool for training deep\nlearning models within a predefined labeling budget. These models have achieved\nperformances comparable to those trained in an offline setting. However, deep\nactive learning faces substantial issues when dealing with classification\ndatasets containing noisy labels. In this literature review, we discuss the\ncurrent state of deep active learning in the presence of label noise,\nhighlighting unique approaches, their strengths, and weaknesses. With the\nrecent success of vision transformers in image classification tasks, we provide\na brief overview and consider how the transformer layers and attention\nmechanisms can be used to enhance diversity, importance, and uncertainty-based\nselection in queries sent to an oracle for labeling. We further propose\nexploring contrastive learning methods to derive good image representations\nthat can aid in selecting high-value samples for labeling in an active learning\nsetting. We also highlight the need for creating unified benchmarks and\nstandardized datasets for deep active learning in the presence of label noise\nfor image classification to promote the reproducibility of research. The review\nconcludes by suggesting avenues for future research in this area.\n", "date": "2023-02-22", "categories": []}
{"id": "2304.13032", "title": "A Unified Active Learning Framework for Annotating Graph Data with\n  Application to Software Source Code Performance Prediction", "abstract": "  Most machine learning and data analytics applications, including performance\nengineering in software systems, require a large number of annotations and\nlabelled data, which might not be available in advance. Acquiring annotations\noften requires significant time, effort, and computational resources, making it\nchallenging. We develop a unified active learning framework specializing in\nsoftware performance prediction to address this task. We begin by parsing the\nsource code to an Abstract Syntax Tree (AST) and augmenting it with data and\ncontrol flow edges. Then, we convert the tree representation of the source code\nto a Flow Augmented-AST graph (FA-AST) representation. Based on the graph\nrepresentation, we construct various graph embeddings (unsupervised and\nsupervised) into a latent space. Given such an embedding, the framework becomes\ntask agnostic since active learning can be performed using any regression\nmethod and query strategy suited for regression. Within this framework, we\ninvestigate the impact of using different levels of information for active and\npassive learning, e.g., partially available labels and unlabeled test data. Our\napproach aims to improve the investment in AI models for different software\nperformance predictions (execution time) based on the structure of the source\ncode. Our real-world experiments reveal that respectable performance can be\nachieved by querying labels for only a small subset of all the data.\n", "date": "2023-04-06", "categories": []}
{"id": "2301.13635", "title": "Active Learning-based Domain Adaptive Localized Polynomial Chaos\n  Expansion", "abstract": "  The paper presents a novel methodology to build surrogate models of\ncomplicated functions by an active learning-based sequential decomposition of\nthe input random space and construction of localized polynomial chaos\nexpansions, referred to as domain adaptive localized polynomial chaos expansion\n(DAL-PCE). The approach utilizes sequential decomposition of the input random\nspace into smaller sub-domains approximated by low-order polynomial expansions.\nThis allows approximation of functions with strong nonlinearties,\ndiscontinuities, and/or singularities. Decomposition of the input random space\nand local approximations alleviates the Gibbs phenomenon for these types of\nproblems and confines error to a very small vicinity near the non-linearity.\nThe global behavior of the surrogate model is therefore significantly better\nthan existing methods as shown in numerical examples. The whole process is\ndriven by an active learning routine that uses the recently proposed $\\Theta$\ncriterion to assess local variance contributions. The proposed approach\nbalances both \\emph{exploitation} of the surrogate model and \\emph{exploration}\nof the input random space and thus leads to efficient and accurate\napproximation of the original mathematical model. The numerical results show\nthe superiority of the DAL-PCE in comparison to (i) a single global polynomial\nchaos expansion and (ii) the recently proposed stochastic spectral embedding\n(SSE) method developed as an accurate surrogate model and which is based on a\nsimilar domain decomposition process. This method represents general framework\nupon which further extensions and refinements can be based, and which can be\ncombined with any technique for non-intrusive polynomial chaos expansion\nconstruction.\n", "date": "2023-01-31", "categories": []}
{"id": "2309.09465", "title": "Active anomaly detection based on deep one-class classification", "abstract": "  Active learning has been utilized as an efficient tool in building anomaly\ndetection models by leveraging expert feedback. In an active learning\nframework, a model queries samples to be labeled by experts and re-trains the\nmodel with the labeled data samples. It unburdens in obtaining annotated\ndatasets while improving anomaly detection performance. However, most of the\nexisting studies focus on helping experts identify as many abnormal data\nsamples as possible, which is a sub-optimal approach for one-class\nclassification-based deep anomaly detection. In this paper, we tackle two\nessential problems of active learning for Deep SVDD: query strategy and\nsemi-supervised learning method. First, rather than solely identifying\nanomalies, our query strategy selects uncertain samples according to an\nadaptive boundary. Second, we apply noise contrastive estimation in training a\none-class classification model to incorporate both labeled normal and abnormal\ndata effectively. We analyze that the proposed query strategy and\nsemi-supervised loss individually improve an active learning process of anomaly\ndetection and further improve when combined together on seven anomaly detection\ndatasets.\n", "date": "2023-09-18", "categories": []}
{"id": "2309.07914", "title": "ALWOD: Active Learning for Weakly-Supervised Object Detection", "abstract": "  Object detection (OD), a crucial vision task, remains challenged by the lack\nof large training datasets with precise object localization labels. In this\nwork, we propose ALWOD, a new framework that addresses this problem by fusing\nactive learning (AL) with weakly and semi-supervised object detection\nparadigms. Because the performance of AL critically depends on the model\ninitialization, we propose a new auxiliary image generator strategy that\nutilizes an extremely small labeled set, coupled with a large weakly tagged set\nof images, as a warm-start for AL. We then propose a new AL acquisition\nfunction, another critical factor in AL success, that leverages the\nstudent-teacher OD pair disagreement and uncertainty to effectively propose the\nmost informative images to annotate. Finally, to complete the AL loop, we\nintroduce a new labeling task delegated to human annotators, based on selection\nand correction of model-proposed detections, which is both rapid and effective\nin labeling the informative images. We demonstrate, across several challenging\nbenchmarks, that ALWOD significantly narrows the gap between the ODs trained on\nfew partially labeled but strategically selected image instances and those that\nrely on the fully-labeled data. Our code is publicly available on\nhttps://github.com/seqam-lab/ALWOD.\n", "date": "2023-09-14", "categories": []}
{"id": "2301.07670", "title": "Active learning for medical image segmentation with stochastic batches", "abstract": "  The performance of learning-based algorithms improves with the amount of\nlabelled data used for training. Yet, manually annotating data is particularly\ndifficult for medical image segmentation tasks because of the limited expert\navailability and intensive manual effort required. To reduce manual labelling,\nactive learning (AL) targets the most informative samples from the unlabelled\nset to annotate and add to the labelled training set. On the one hand, most\nactive learning works have focused on the classification or limited\nsegmentation of natural images, despite active learning being highly desirable\nin the difficult task of medical image segmentation. On the other hand,\nuncertainty-based AL approaches notoriously offer sub-optimal batch-query\nstrategies, while diversity-based methods tend to be computationally expensive.\nOver and above methodological hurdles, random sampling has proven an extremely\ndifficult baseline to outperform when varying learning and sampling conditions.\nThis work aims to take advantage of the diversity and speed offered by random\nsampling to improve the selection of uncertainty-based AL methods for\nsegmenting medical images. More specifically, we propose to compute uncertainty\nat the level of batches instead of samples through an original use of\nstochastic batches (SB) during sampling in AL. Stochastic batch querying is a\nsimple and effective add-on that can be used on top of any uncertainty-based\nmetric. Extensive experiments on two medical image segmentation datasets show\nthat our strategy consistently improves conventional uncertainty-based sampling\nmethods. Our method can hence act as a strong baseline for medical image\nsegmentation. The code is available on:\nhttps://github.com/Minimel/StochasticBatchAL.git.\n", "date": "2023-01-18", "categories": []}
{"id": "2309.05477", "title": "Learning Objective-Specific Active Learning Strategies with Attentive\n  Neural Processes", "abstract": "  Pool-based active learning (AL) is a promising technology for increasing\ndata-efficiency of machine learning models. However, surveys show that\nperformance of recent AL methods is very sensitive to the choice of dataset and\ntraining setting, making them unsuitable for general application. In order to\ntackle this problem, the field Learning Active Learning (LAL) suggests to learn\nthe active learning strategy itself, allowing it to adapt to the given setting.\nIn this work, we propose a novel LAL method for classification that exploits\nsymmetry and independence properties of the active learning problem with an\nAttentive Conditional Neural Process model. Our approach is based on learning\nfrom a myopic oracle, which gives our model the ability to adapt to\nnon-standard objectives, such as those that do not equally weight the error on\nall data points. We experimentally verify that our Neural Process model\noutperforms a variety of baselines in these settings. Finally, our experiments\nshow that our model exhibits a tendency towards improved stability to changing\ndatasets. However, performance is sensitive to choice of classifier and more\nwork is necessary to reduce the performance the gap with the myopic oracle and\nto improve scalability. We present our work as a proof-of-concept for LAL on\nnonstandard objectives and hope our analysis and modelling considerations\ninspire future LAL work.\n", "date": "2023-09-11", "categories": []}
{"id": "2302.03347", "title": "An Informative Path Planning Framework for Active Learning in UAV-based\n  Semantic Mapping", "abstract": "  Unmanned aerial vehicles (UAVs) are frequently used for aerial mapping and\ngeneral monitoring tasks. Recent progress in deep learning enabled automated\nsemantic segmentation of imagery to facilitate the interpretation of\nlarge-scale complex environments. Commonly used supervised deep learning for\nsegmentation relies on large amounts of pixel-wise labelled data, which is\ntedious and costly to annotate. The domain-specific visual appearance of aerial\nenvironments often prevents the usage of models pre-trained on publicly\navailable datasets. To address this, we propose a novel general planning\nframework for UAVs to autonomously acquire informative training images for\nmodel re-training. We leverage multiple acquisition functions and fuse them\ninto probabilistic terrain maps. Our framework combines the mapped acquisition\nfunction information into the UAV's planning objectives. In this way, the UAV\nadaptively acquires informative aerial images to be manually labelled for model\nre-training. Experimental results on real-world data and in a photorealistic\nsimulation show that our framework maximises model performance and drastically\nreduces labelling efforts. Our map-based planners outperform state-of-the-art\nlocal planning.\n", "date": "2023-02-07", "categories": []}
{"id": "2308.14781", "title": "Conflict-Aware Active Automata Learning (Extended Version)", "abstract": "  Active automata learning algorithms cannot easily handle conflict in the\nobservation data (different outputs observed for the same inputs). This\ninherent inability to recover after a conflict impairs their effective\napplicability in scenarios where noise is present or the system under learning\nis mutating.\n  We propose the Conflict-Aware Active Automata Learning (C3AL) framework to\nenable handling conflicting information during the learning process. The core\nidea is to consider the so-called observation tree as a first-class citizen in\nthe learning process. Though this idea is explored in recent work, we take it\nto its full effect by enabling its use with any existing learner and minimizing\nthe number of tests performed on the system under learning, specially in the\nface of conflicts. We evaluate C3AL in a large set of benchmarks, covering over\n30 different realistic targets, and over 18,000 different scenarios. The\nresults of the evaluation show that C3AL is a suitable alternative framework\nfor closed-box learning that can better handle noise and mutations.\n", "date": "2023-08-28", "categories": []}
{"id": "2309.02476", "title": "Optimal Sample Selection Through Uncertainty Estimation and Its\n  Application in Deep Learning", "abstract": "  Modern deep learning heavily relies on large labeled datasets, which often\ncomse with high costs in terms of both manual labeling and computational\nresources. To mitigate these challenges, researchers have explored the use of\ninformative subset selection techniques, including coreset selection and active\nlearning. Specifically, coreset selection involves sampling data with both\ninput ($\\bx$) and output ($\\by$), active learning focuses solely on the input\ndata ($\\bx$).\n  In this study, we present a theoretically optimal solution for addressing\nboth coreset selection and active learning within the context of linear softmax\nregression. Our proposed method, COPS (unCertainty based OPtimal Sub-sampling),\nis designed to minimize the expected loss of a model trained on subsampled\ndata. Unlike existing approaches that rely on explicit calculations of the\ninverse covariance matrix, which are not easily applicable to deep learning\nscenarios, COPS leverages the model's logits to estimate the sampling ratio.\nThis sampling ratio is closely associated with model uncertainty and can be\neffectively applied to deep learning tasks. Furthermore, we address the\nchallenge of model sensitivity to misspecification by incorporating a\ndown-weighting approach for low-density samples, drawing inspiration from\nprevious works.\n  To assess the effectiveness of our proposed method, we conducted extensive\nempirical experiments using deep neural networks on benchmark datasets. The\nresults consistently showcase the superior performance of COPS compared to\nbaseline methods, reaffirming its efficacy.\n", "date": "2023-09-05", "categories": []}
{"id": "2309.02818", "title": "Combining Thermodynamics-based Model of the Centrifugal Compressors and\n  Active Machine Learning for Enhanced Industrial Design Optimization", "abstract": "  The design process of centrifugal compressors requires applying an\noptimization process which is computationally expensive due to complex\nanalytical equations underlying the compressor's dynamical equations. Although\nthe regression surrogate models could drastically reduce the computational cost\nof such a process, the major challenge is the scarcity of data for training the\nsurrogate model. Aiming to strategically exploit the labeled samples, we\npropose the Active-CompDesign framework in which we combine a\nthermodynamics-based compressor model (i.e., our internal software for\ncompressor design) and Gaussian Process-based surrogate model within a\ndeployable Active Learning (AL) setting. We first conduct experiments in an\noffline setting and further, extend it to an online AL framework where a\nreal-time interaction with the thermodynamics-based compressor's model allows\nthe deployment in production. ActiveCompDesign shows a significant performance\nimprovement in surrogate modeling by leveraging on uncertainty-based query\nfunction of samples within the AL framework with respect to the random\nselection of data points. Moreover, our framework in production has reduced the\ntotal computational time of compressor's design optimization to around 46%\nfaster than relying on the internal thermodynamics-based simulator, achieving\nthe same performance.\n", "date": "2023-09-06", "categories": []}
{"id": "2309.01612", "title": "On the Query Strategies for Efficient Online Active Distillation", "abstract": "  Deep Learning (DL) requires lots of time and data, resulting in high\ncomputational demands. Recently, researchers employ Active Learning (AL) and\nonline distillation to enhance training efficiency and real-time model\nadaptation. This paper evaluates a set of query strategies to achieve the best\ntraining results. It focuses on Human Pose Estimation (HPE) applications,\nassessing the impact of selected frames during training using two approaches: a\nclassical offline method and a online evaluation through a continual learning\napproach employing knowledge distillation, on a popular state-of-the-art HPE\ndataset. The paper demonstrates the possibility of enabling training at the\nedge lightweight models, adapting them effectively to new contexts in\nreal-time.\n", "date": "2023-09-04", "categories": []}
{"id": "2308.12740", "title": "Human Comprehensible Active Learning of Genome-Scale Metabolic Networks", "abstract": "  An important application of Synthetic Biology is the engineering of the host\ncell system to yield useful products. However, an increase in the scale of the\nhost system leads to huge design space and requires a large number of\nvalidation trials with high experimental costs. A comprehensible machine\nlearning approach that efficiently explores the hypothesis space and guides\nexperimental design is urgently needed for the Design-Build-Test-Learn (DBTL)\ncycle of the host cell system. We introduce a novel machine learning framework\nILP-iML1515 based on Inductive Logic Programming (ILP) that performs abductive\nlogical reasoning and actively learns from training examples. In contrast to\nnumerical models, ILP-iML1515 is built on comprehensible logical\nrepresentations of a genome-scale metabolic model and can update the model by\nlearning new logical structures from auxotrophic mutant trials. The ILP-iML1515\nframework 1) allows high-throughput simulations and 2) actively selects\nexperiments that reduce the experimental cost of learning gene functions in\ncomparison to randomly selected experiments.\n", "date": "2023-08-24", "categories": []}
{"id": "2309.00356", "title": "Explainable Active Learning for Preference Elicitation", "abstract": "  Gaining insights into the preferences of new users and subsequently\npersonalizing recommendations necessitate managing user interactions\nintelligently, namely, posing pertinent questions to elicit valuable\ninformation effectively. In this study, our focus is on a specific scenario of\nthe cold-start problem, where the recommendation system lacks adequate user\npresence or access to other users' data is restricted, obstructing employing\nuser profiling methods utilizing existing data in the system. We employ Active\nLearning (AL) to solve the addressed problem with the objective of maximizing\ninformation acquisition with minimal user effort. AL operates for selecting\ninformative data from a large unlabeled set to inquire an oracle to label them\nand eventually updating a machine learning (ML) model. We operate AL in an\nintegrated process of unsupervised, semi-supervised, and supervised ML within\nan explanatory preference elicitation process. It harvests user feedback (given\nfor the system's explanations on the presented items) over informative samples\nto update an underlying ML model estimating user preferences. The designed user\ninteraction facilitates personalizing the system by incorporating user feedback\ninto the ML model and also enhances user trust by refining the system's\nexplanations on recommendations. We implement the proposed preference\nelicitation methodology for food recommendation. We conducted human experiments\nto assess its efficacy in the short term and also experimented with several AL\nstrategies over synthetic user profiles that we created for two food datasets,\naiming for long-term performance analysis. The experimental results demonstrate\nthe efficiency of the proposed preference elicitation with limited user-labeled\ndata while also enhancing user trust through accurate explanations.\n", "date": "2023-09-01", "categories": []}
{"id": "2308.14904", "title": "Maturity-Aware Active Learning for Semantic Segmentation with\n  Hierarchically-Adaptive Sample Assessment", "abstract": "  Active Learning (AL) for semantic segmentation is challenging due to heavy\nclass imbalance and different ways of defining \"sample\" (pixels, areas, etc.),\nleaving the interpretation of the data distribution ambiguous. We propose\n\"Maturity-Aware Distribution Breakdown-based Active Learning'' (MADBAL), an AL\nmethod that benefits from a hierarchical approach to define a multiview data\ndistribution, which takes into account the different \"sample\" definitions\njointly, hence able to select the most impactful segmentation pixels with\ncomprehensive understanding. MADBAL also features a novel uncertainty\nformulation, where AL supporting modules are included to sense the features'\nmaturity whose weighted influence continuously contributes to the uncertainty\ndetection. In this way, MADBAL makes significant performance leaps even in the\nearly AL stage, hence reducing the training burden significantly. It\noutperforms state-of-the-art methods on Cityscapes and PASCAL VOC datasets as\nverified in our extensive experiments.\n", "date": "2023-08-28", "categories": []}
{"id": "2211.10773", "title": "A Two-Stage Active Learning Algorithm for $k$-Nearest Neighbors", "abstract": "  $k$-nearest neighbor classification is a popular non-parametric method\nbecause of desirable properties like automatic adaption to distributional scale\nchanges. Unfortunately, it has thus far proved difficult to design active\nlearning strategies for the training of local voting-based classifiers that\nnaturally retain these desirable properties, and hence active learning\nstrategies for $k$-nearest neighbor classification have been conspicuously\nmissing from the literature. In this work, we introduce a simple and intuitive\nactive learning algorithm for the training of $k$-nearest neighbor classifiers,\nthe first in the literature which retains the concept of the $k$-nearest\nneighbor vote at prediction time. We provide consistency guarantees for a\nmodified $k$-nearest neighbors classifier trained on samples acquired via our\nscheme, and show that when the conditional probability function\n$\\mathbb{P}(Y=y|X=x)$ is sufficiently smooth and the Tsybakov noise condition\nholds, our actively trained classifiers converge to the Bayes optimal\nclassifier at a faster asymptotic rate than passively trained $k$-nearest\nneighbor classifiers.\n", "date": "2022-11-19", "categories": []}
{"id": "2303.16817", "title": "Adaptive Superpixel for Active Learning in Semantic Segmentation", "abstract": "  Learning semantic segmentation requires pixel-wise annotations, which can be\ntime-consuming and expensive. To reduce the annotation cost, we propose a\nsuperpixel-based active learning (AL) framework, which collects a dominant\nlabel per superpixel instead. To be specific, it consists of adaptive\nsuperpixel and sieving mechanisms, fully dedicated to AL. At each round of AL,\nwe adaptively merge neighboring pixels of similar learned features into\nsuperpixels. We then query a selected subset of these superpixels using an\nacquisition function assuming no uniform superpixel size. This approach is more\nefficient than existing methods, which rely only on innate features such as RGB\ncolor and assume uniform superpixel sizes. Obtaining a dominant label per\nsuperpixel drastically reduces annotators' burden as it requires fewer clicks.\nHowever, it inevitably introduces noisy annotations due to mismatches between\nsuperpixel and ground truth segmentation. To address this issue, we further\ndevise a sieving mechanism that identifies and excludes potentially noisy\nannotations from learning. Our experiments on both Cityscapes and PASCAL VOC\ndatasets demonstrate the efficacy of adaptive superpixel and sieving\nmechanisms.\n", "date": "2023-03-29", "categories": []}
{"id": "2306.08306", "title": "Towards Balanced Active Learning for Multimodal Classification", "abstract": "  Training multimodal networks requires a vast amount of data due to their\nlarger parameter space compared to unimodal networks. Active learning is a\nwidely used technique for reducing data annotation costs by selecting only\nthose samples that could contribute to improving model performance. However,\ncurrent active learning strategies are mostly designed for unimodal tasks, and\nwhen applied to multimodal data, they often result in biased sample selection\nfrom the dominant modality. This unfairness hinders balanced multimodal\nlearning, which is crucial for achieving optimal performance. To address this\nissue, we propose three guidelines for designing a more balanced multimodal\nactive learning strategy. Following these guidelines, a novel approach is\nproposed to achieve more fair data selection by modulating the gradient\nembedding with the dominance degree among modalities. Our studies demonstrate\nthat the proposed method achieves more balanced multimodal learning by avoiding\ngreedy sample selection from the dominant modality. Our approach outperforms\nexisting active learning strategies on a variety of multimodal classification\ntasks. Overall, our work highlights the importance of balancing sample\nselection in multimodal active learning and provides a practical solution for\nachieving more balanced active learning for multimodal classification.\n", "date": "2023-06-14", "categories": []}
{"id": "2308.10571", "title": "Overcoming Overconfidence for Active Learning", "abstract": "  It is not an exaggeration to say that the recent progress in artificial\nintelligence technology depends on large-scale and high-quality data.\nSimultaneously, a prevalent issue exists everywhere: the budget for data\nlabeling is constrained. Active learning is a prominent approach for addressing\nthis issue, where valuable data for labeling is selected through a model and\nutilized to iteratively adjust the model. However, due to the limited amount of\ndata in each iteration, the model is vulnerable to bias; thus, it is more\nlikely to yield overconfident predictions. In this paper, we present two novel\nmethods to address the problem of overconfidence that arises in the active\nlearning scenario. The first is an augmentation strategy named\nCross-Mix-and-Mix (CMaM), which aims to calibrate the model by expanding the\nlimited training distribution. The second is a selection strategy named Ranked\nMargin Sampling (RankedMS), which prevents choosing data that leads to overly\nconfident predictions. Through various experiments and analyses, we are able to\ndemonstrate that our proposals facilitate efficient data selection by\nalleviating overconfidence, even though they are readily applicable.\n", "date": "2023-08-21", "categories": []}
{"id": "2308.08823", "title": "Mitigating Semantic Confusion from Hostile Neighborhood for Graph Active\n  Learning", "abstract": "  Graph Active Learning (GAL), which aims to find the most informative nodes in\ngraphs for annotation to maximize the Graph Neural Networks (GNNs) performance,\nhas attracted many research efforts but remains non-trivial challenges. One\nmajor challenge is that existing GAL strategies may introduce semantic\nconfusion to the selected training set, particularly when graphs are noisy.\nSpecifically, most existing methods assume all aggregating features to be\nhelpful, ignoring the semantically negative effect between inter-class edges\nunder the message-passing mechanism. In this work, we present Semantic-aware\nActive learning framework for Graphs (SAG) to mitigate the semantic confusion\nproblem. Pairwise similarities and dissimilarities of nodes with semantic\nfeatures are introduced to jointly evaluate the node influence. A new\nprototype-based criterion and query policy are also designed to maintain\ndiversity and class balance of the selected nodes, respectively. Extensive\nexperiments on the public benchmark graphs and a real-world financial dataset\ndemonstrate that SAG significantly improves node classification performances\nand consistently outperforms previous methods. Moreover, comprehensive analysis\nand ablation study also verify the effectiveness of the proposed framework.\n", "date": "2023-08-17", "categories": []}
{"id": "2209.04744", "title": "Active Learning for Optimal Intervention Design in Causal Models", "abstract": "  Sequential experimental design to discover interventions that achieve a\ndesired outcome is a key problem in various domains including science,\nengineering and public policy. When the space of possible interventions is\nlarge, making an exhaustive search infeasible, experimental design strategies\nare needed. In this context, encoding the causal relationships between the\nvariables, and thus the effect of interventions on the system, is critical for\nidentifying desirable interventions more efficiently. Here, we develop a causal\nactive learning strategy to identify interventions that are optimal, as\nmeasured by the discrepancy between the post-interventional mean of the\ndistribution and a desired target mean. The approach employs a Bayesian update\nfor the causal model and prioritizes interventions using a carefully designed,\ncausally informed acquisition function. This acquisition function is evaluated\nin closed form, allowing for fast optimization. The resulting algorithms are\ntheoretically grounded with information-theoretic bounds and provable\nconsistency results for linear causal models with known causal graph. We apply\nour approach to both synthetic data and single-cell transcriptomic data from\nPerturb-CITE-seq experiments to identify optimal perturbations that induce a\nspecific cell state transition. The causally informed acquisition function\ngenerally outperforms existing criteria allowing for optimal intervention\ndesign with fewer but carefully selected samples.\n", "date": "2022-09-10", "categories": []}
{"id": "2304.04137", "title": "RD-DPP: Rate-Distortion Theory Meets Determinantal Point Process to\n  Diversify Learning Data Samples", "abstract": "  In some practical learning tasks, such as traffic video analysis, the number\nof available training samples is restricted by different factors, such as\nlimited communication bandwidth and computation power. Determinantal Point\nProcess (DPP) is a common method for selecting the most diverse samples to\nenhance learning quality. However, the number of selected samples is restricted\nto the rank of the kernel matrix implied by the dimensionality of data samples.\nSecondly, it is not easily customizable to different learning tasks. In this\npaper, we propose a new way of measuring task-oriented diversity based on the\nRate-Distortion (RD) theory, appropriate for multi-level classification. To\nthis end, we establish a fundamental relationship between DPP and RD theory. We\nobserve that the upper bound of the diversity of data selected by DPP has a\nuniversal trend of $\\textit{phase transition}$, which suggests that DPP is\nbeneficial only at the beginning of sample accumulation. This led to the design\nof a bi-modal method, where RD-DPP is used in the first mode to select initial\ndata samples, then classification inconsistency (as an uncertainty measure) is\nused to select the subsequent samples in the second mode. This phase transition\nsolves the limitation to the rank of the similarity matrix. Applying our method\nto six different datasets and five benchmark models suggests that our method\nconsistently outperforms random selection, DPP-based methods, and alternatives\nlike uncertainty-based and coreset methods under all sampling budgets, while\nexhibiting high generalizability to different learning tasks.\n", "date": "2023-04-09", "categories": []}
{"id": "2308.08003", "title": "BI-LAVA: Biocuration with Hierarchical Image Labeling through Active\n  Learning and Visual Analysis", "abstract": "  In the biomedical domain, taxonomies organize the acquisition modalities of\nscientific images in hierarchical structures. Such taxonomies leverage large\nsets of correct image labels and provide essential information about the\nimportance of a scientific publication, which could then be used in biocuration\ntasks. However, the hierarchical nature of the labels, the overhead of\nprocessing images, the absence or incompleteness of labeled data, and the\nexpertise required to label this type of data impede the creation of useful\ndatasets for biocuration. From a multi-year collaboration with biocurators and\ntext-mining researchers, we derive an iterative visual analytics and active\nlearning strategy to address these challenges. We implement this strategy in a\nsystem called BI-LAVA Biocuration with Hierarchical Image Labeling through\nActive Learning and Visual Analysis. BI-LAVA leverages a small set of image\nlabels, a hierarchical set of image classifiers, and active learning to help\nmodel builders deal with incomplete ground-truth labels, target a hierarchical\ntaxonomy of image modalities, and classify a large pool of unlabeled images.\nBI-LAVA's front end uses custom encodings to represent data distributions,\ntaxonomies, image projections, and neighborhoods of image thumbnails, which\nhelp model builders explore an unfamiliar image dataset and taxonomy and\ncorrect and generate labels. An evaluation with machine learning practitioners\nshows that our mixed human-machine approach successfully supports domain\nexperts in understanding the characteristics of classes within the taxonomy, as\nwell as validating and improving data quality in labeled and unlabeled\ncollections.\n", "date": "2023-08-15", "categories": []}
{"id": "2308.08224", "title": "How To Overcome Confirmation Bias in Semi-Supervised Image\n  Classification By Active Learning", "abstract": "  Do we need active learning? The rise of strong deep semi-supervised methods\nraises doubt about the usability of active learning in limited labeled data\nsettings. This is caused by results showing that combining semi-supervised\nlearning (SSL) methods with a random selection for labeling can outperform\nexisting active learning (AL) techniques. However, these results are obtained\nfrom experiments on well-established benchmark datasets that can overestimate\nthe external validity. However, the literature lacks sufficient research on the\nperformance of active semi-supervised learning methods in realistic data\nscenarios, leaving a notable gap in our understanding. Therefore we present\nthree data challenges common in real-world applications: between-class\nimbalance, within-class imbalance, and between-class similarity. These\nchallenges can hurt SSL performance due to confirmation bias. We conduct\nexperiments with SSL and AL on simulated data challenges and find that random\nsampling does not mitigate confirmation bias and, in some cases, leads to worse\nperformance than supervised learning. In contrast, we demonstrate that AL can\novercome confirmation bias in SSL in these realistic settings. Our results\nprovide insights into the potential of combining active and semi-supervised\nlearning in the presence of common real-world challenges, which is a promising\ndirection for robust methods when learning with limited labeled data in\nreal-world applications.\n", "date": "2023-08-16", "categories": []}
{"id": "2303.06457", "title": "Active Visual Exploration Based on Attention-Map Entropy", "abstract": "  Active visual exploration addresses the issue of limited sensor capabilities\nin real-world scenarios, where successive observations are actively chosen\nbased on the environment. To tackle this problem, we introduce a new technique\ncalled Attention-Map Entropy (AME). It leverages the internal uncertainty of\nthe transformer-based model to determine the most informative observations. In\ncontrast to existing solutions, it does not require additional loss components,\nwhich simplifies the training. Through experiments, which also mimic\nretina-like sensors, we show that such simplified training significantly\nimproves the performance of reconstruction, segmentation and classification on\npublicly available datasets.\n", "date": "2023-03-11", "categories": []}
{"id": "2308.02537", "title": "ALE: A Simulation-Based Active Learning Evaluation Framework for the\n  Parameter-Driven Comparison of Query Strategies for NLP", "abstract": "  Supervised machine learning and deep learning require a large amount of\nlabeled data, which data scientists obtain in a manual, and time-consuming\nannotation process. To mitigate this challenge, Active Learning (AL) proposes\npromising data points to annotators they annotate next instead of a subsequent\nor random sample. This method is supposed to save annotation effort while\nmaintaining model performance. However, practitioners face many AL strategies\nfor different tasks and need an empirical basis to choose between them. Surveys\ncategorize AL strategies into taxonomies without performance indications.\nPresentations of novel AL strategies compare the performance to a small subset\nof strategies. Our contribution addresses the empirical basis by introducing a\nreproducible active learning evaluation (ALE) framework for the comparative\nevaluation of AL strategies in NLP. The framework allows the implementation of\nAL strategies with low effort and a fair data-driven comparison through\ndefining and tracking experiment parameters (e.g., initial dataset size, number\nof data points per query step, and the budget). ALE helps practitioners to make\nmore informed decisions, and researchers can focus on developing new, effective\nAL strategies and deriving best practices for specific use cases. With best\npractices, practitioners can lower their annotation costs. We present a case\nstudy to illustrate how to use the framework.\n", "date": "2023-08-01", "categories": []}
{"id": "2305.18905", "title": "atTRACTive: Semi-automatic white matter tract segmentation using active\n  learning", "abstract": "  Accurately identifying white matter tracts in medical images is essential for\nvarious applications, including surgery planning and tract-specific analysis.\nSupervised machine learning models have reached state-of-the-art solving this\ntask automatically. However, these models are primarily trained on healthy\nsubjects and struggle with strong anatomical aberrations, e.g. caused by brain\ntumors. This limitation makes them unsuitable for tasks such as preoperative\nplanning, wherefore time-consuming and challenging manual delineation of the\ntarget tract is typically employed. We propose semi-automatic entropy-based\nactive learning for quick and intuitive segmentation of white matter tracts\nfrom whole-brain tractography consisting of millions of streamlines. The method\nis evaluated on 21 openly available healthy subjects from the Human Connectome\nProject and an internal dataset of ten neurosurgical cases. With only a few\nannotations, the proposed approach enables segmenting tracts on tumor cases\ncomparable to healthy subjects (dice=0.71), while the performance of automatic\nmethods, like TractSeg dropped substantially (dice=0.34) in comparison to\nhealthy subjects. The method is implemented as a prototype named atTRACTive in\nthe freely available software MITK Diffusion. Manual experiments on tumor data\nshowed higher efficiency due to lower segmentation times compared to\ntraditional ROI-based segmentation.\n", "date": "2023-05-30", "categories": []}
{"id": "2303.15823", "title": "Automated wildlife image classification: An active learning tool for\n  ecological applications", "abstract": "  Wildlife camera trap images are being used extensively to investigate animal\nabundance, habitat associations, and behavior, which is complicated by the fact\nthat experts must first classify the images manually. Artificial intelligence\nsystems can take over this task but usually need a large number of\nalready-labeled training images to achieve sufficient performance. This\nrequirement necessitates human expert labor and poses a particular challenge\nfor projects with few cameras or short durations. We propose a label-efficient\nlearning strategy that enables researchers with small or medium-sized image\ndatabases to leverage the potential of modern machine learning, thus freeing\ncrucial resources for subsequent analyses.\n  Our methodological proposal is two-fold: (1) We improve current strategies of\ncombining object detection and image classification by tuning the\nhyperparameters of both models. (2) We provide an active learning (AL) system\nthat allows training deep learning models very efficiently in terms of required\nhuman-labeled training images. We supply a software package that enables\nresearchers to use these methods directly and thereby ensure the broad\napplicability of the proposed framework in ecological practice.\n  We show that our tuning strategy improves predictive performance. We\ndemonstrate how the AL pipeline reduces the amount of pre-labeled data needed\nto achieve a specific predictive performance and that it is especially valuable\nfor improving out-of-sample predictive performance.\n  We conclude that the combination of tuning and AL increases predictive\nperformance substantially. Furthermore, we argue that our work can broadly\nimpact the community through the ready-to-use software package provided.\nFinally, the publication of our models tailored to European wildlife data\nenriches existing model bases mostly trained on data from Africa and North\nAmerica.\n", "date": "2023-03-28", "categories": []}
{"id": "2211.00928", "title": "Neural Active Learning on Heteroskedastic Distributions", "abstract": "  Models that can actively seek out the best quality training data hold the\npromise of more accurate, adaptable, and efficient machine learning. Active\nlearning techniques often tend to prefer examples that are the most difficult\nto classify. While this works well on homogeneous datasets, we find that it can\nlead to catastrophic failures when performed on multiple distributions with\ndifferent degrees of label noise or heteroskedasticity. These active learning\nalgorithms strongly prefer to draw from the distribution with more noise, even\nif their examples have no informative structure (such as solid color images\nwith random labels). To this end, we demonstrate the catastrophic failure of\nthese active learning algorithms on heteroskedastic distributions and propose a\nfine-tuning-based approach to mitigate these failures. Further, we propose a\nnew algorithm that incorporates a model difference scoring function for each\ndata point to filter out the noisy examples and sample clean examples that\nmaximize accuracy, outperforming the existing active learning techniques on the\nheteroskedastic datasets. We hope these observations and techniques are\nimmediately helpful to practitioners and can help to challenge common\nassumptions in the design of active learning algorithms.\n", "date": "2022-11-02", "categories": []}
{"id": "2307.12004", "title": "COLosSAL: A Benchmark for Cold-start Active Learning for 3D Medical\n  Image Segmentation", "abstract": "  Medical image segmentation is a critical task in medical image analysis. In\nrecent years, deep learning based approaches have shown exceptional performance\nwhen trained on a fully-annotated dataset. However, data annotation is often a\nsignificant bottleneck, especially for 3D medical images. Active learning (AL)\nis a promising solution for efficient annotation but requires an initial set of\nlabeled samples to start active selection. When the entire data pool is\nunlabeled, how do we select the samples to annotate as our initial set? This is\nalso known as the cold-start AL, which permits only one chance to request\nannotations from experts without access to previously annotated data.\nCold-start AL is highly relevant in many practical scenarios but has been\nunder-explored, especially for 3D medical segmentation tasks requiring\nsubstantial annotation effort. In this paper, we present a benchmark named\nCOLosSAL by evaluating six cold-start AL strategies on five 3D medical image\nsegmentation tasks from the public Medical Segmentation Decathlon collection.\nWe perform a thorough performance analysis and explore important open questions\nfor cold-start AL, such as the impact of budget on different strategies. Our\nresults show that cold-start AL is still an unsolved problem for 3D\nsegmentation tasks but some important trends have been observed. The code\nrepository, data partitions, and baseline results for the complete benchmark\nare publicly available at https://github.com/MedICL-VU/COLosSAL.\n", "date": "2023-07-22", "categories": []}
{"id": "2307.11209", "title": "Clinical Trial Active Learning", "abstract": "  This paper presents a novel approach to active learning that takes into\naccount the non-independent and identically distributed (non-i.i.d.) structure\nof a clinical trial setting. There exists two types of clinical trials:\nretrospective and prospective. Retrospective clinical trials analyze data after\ntreatment has been performed; prospective clinical trials collect data as\ntreatment is ongoing. Typically, active learning approaches assume the dataset\nis i.i.d. when selecting training samples; however, in the case of clinical\ntrials, treatment results in a dependency between the data collected at the\ncurrent and past visits. Thus, we propose prospective active learning to\novercome the limitations present in traditional active learning methods and\napply it to disease detection in optical coherence tomography (OCT) images,\nwhere we condition on the time an image was collected to enforce the i.i.d.\nassumption. We compare our proposed method to the traditional active learning\nparadigm, which we refer to as retrospective in nature. We demonstrate that\nprospective active learning outperforms retrospective active learning in two\ndifferent types of test settings.\n", "date": "2023-07-20", "categories": []}
{"id": "2307.11227", "title": "UP-DP: Unsupervised Prompt Learning for Data Pre-Selection with\n  Vision-Language Models", "abstract": "  In this study, we investigate the task of data pre-selection, which aims to\nselect instances for labeling from an unlabeled dataset through a single pass,\nthereby optimizing performance for undefined downstream tasks with a limited\nannotation budget. Previous approaches to data pre-selection relied solely on\nvisual features extracted from foundation models, such as CLIP and BLIP-2, but\nlargely ignored the powerfulness of text features. In this work, we argue that,\nwith proper design, the joint feature space of both vision and text can yield a\nbetter representation for data pre-selection. To this end, we introduce UP-DP,\na simple yet effective unsupervised prompt learning approach that adapts\nvision-language models, like BLIP-2, for data pre-selection. Specifically, with\nthe BLIP-2 parameters frozen, we train text prompts to extract the joint\nfeatures with improved representation, ensuring a diverse cluster structure\nthat covers the entire dataset. We extensively compare our method with the\nstate-of-the-art using seven benchmark datasets in different settings,\nachieving up to a performance gain of 20%. Interestingly, the prompts learned\nfrom one dataset demonstrate significant generalizability and can be applied\ndirectly to enhance the feature extraction of BLIP-2 from other datasets. To\nthe best of our knowledge, UP-DP is the first work to incorporate unsupervised\nprompt learning in a vision-language model for data pre-selection.\n", "date": "2023-07-20", "categories": []}
{"id": "2307.10495", "title": "Novel Batch Active Learning Approach and Its Application to Synthetic\n  Aperture Radar Datasets", "abstract": "  Active learning improves the performance of machine learning methods by\njudiciously selecting a limited number of unlabeled data points to query for\nlabels, with the aim of maximally improving the underlying classifier's\nperformance. Recent gains have been made using sequential active learning for\nsynthetic aperture radar (SAR) data arXiv:2204.00005. In each iteration,\nsequential active learning selects a query set of size one while batch active\nlearning selects a query set of multiple datapoints. While batch active\nlearning methods exhibit greater efficiency, the challenge lies in maintaining\nmodel accuracy relative to sequential active learning methods. We developed a\nnovel, two-part approach for batch active learning: Dijkstra's Annulus Core-Set\n(DAC) for core-set generation and LocalMax for batch sampling. The batch active\nlearning process that combines DAC and LocalMax achieves nearly identical\naccuracy as sequential active learning but is more efficient, proportional to\nthe batch size. As an application, a pipeline is built based on transfer\nlearning feature embedding, graph learning, DAC, and LocalMax to classify the\nFUSAR-Ship and OpenSARShip datasets. Our pipeline outperforms the\nstate-of-the-art CNN-based methods.\n", "date": "2023-07-19", "categories": []}
{"id": "2210.13601", "title": "Active Learning for Single Neuron Models with Lipschitz Non-Linearities", "abstract": "  We consider the problem of active learning for single neuron models, also\nsometimes called ``ridge functions'', in the agnostic setting (under\nadversarial label noise). Such models have been shown to be broadly effective\nin modeling physical phenomena, and for constructing surrogate data-driven\nmodels for partial differential equations.\n  Surprisingly, we show that for a single neuron model with any Lipschitz\nnon-linearity (such as the ReLU, sigmoid, absolute value, low-degree\npolynomial, among others), strong provable approximation guarantees can be\nobtained using a well-known active learning strategy for fitting \\emph{linear\nfunctions} in the agnostic setting. % -- i.e. for the case when there is no\nnon-linearity. Namely, we can collect samples via statistical \\emph{leverage\nscore sampling}, which has been shown to be near-optimal in other active\nlearning scenarios. We support our theoretical results with empirical\nsimulations showing that our proposed active learning strategy based on\nleverage score sampling outperforms (ordinary) uniform sampling when fitting\nsingle neuron models.\n", "date": "2022-10-24", "categories": []}
{"id": "2302.00422", "title": "Robust online active learning", "abstract": "  In many industrial applications, obtaining labeled observations is not\nstraightforward as it often requires the intervention of human experts or the\nuse of expensive testing equipment. In these circumstances, active learning can\nbe highly beneficial in suggesting the most informative data points to be used\nwhen fitting a model. Reducing the number of observations needed for model\ndevelopment alleviates both the computational burden required for training and\nthe operational expenses related to labeling. Online active learning, in\nparticular, is useful in high-volume production processes where the decision\nabout the acquisition of the label for a data point needs to be taken within an\nextremely short time frame. However, despite the recent efforts to develop\nonline active learning strategies, the behavior of these methods in the\npresence of outliers has not been thoroughly examined. In this work, we\ninvestigate the performance of online active linear regression in contaminated\ndata streams. Our study shows that the currently available query strategies are\nprone to sample outliers, whose inclusion in the training set eventually\ndegrades the predictive performance of the models. To address this issue, we\npropose a solution that bounds the search area of a conditional D-optimal\nalgorithm and uses a robust estimator. Our approach strikes a balance between\nexploring unseen regions of the input space and protecting against outliers.\nThrough numerical simulations, we show that the proposed method is effective in\nimproving the performance of online active learning in the presence of\noutliers, thus expanding the potential applications of this powerful tool.\n", "date": "2023-02-01", "categories": []}
{"id": "2307.08782", "title": "Unsupervised Learning of Distributional Properties can Supplement Human\n  Labeling and Increase Active Learning Efficiency in Anomaly Detection", "abstract": "  Exfiltration of data via email is a serious cybersecurity threat for many\norganizations. Detecting data exfiltration (anomaly) patterns typically\nrequires labeling, most often done by a human annotator, to reduce the high\nnumber of false alarms. Active Learning (AL) is a promising approach for\nlabeling data efficiently, but it needs to choose an efficient order in which\ncases are to be labeled, and there are uncertainties as to what scoring\nprocedure should be used to prioritize cases for labeling, especially when\ndetecting rare cases of interest is crucial. We propose an adaptive AL sampling\nstrategy that leverages the underlying prior data distribution, as well as\nmodel uncertainty, to produce batches of cases to be labeled that contain\ninstances of rare anomalies. We show that (1) the classifier benefits from a\nbatch of representative and informative instances of both normal and anomalous\nexamples, (2) unsupervised anomaly detection plays a useful role in building\nthe classifier in the early stages of training when relatively little labeling\nhas been done thus far. Our approach to AL for anomaly detection outperformed\nexisting AL approaches on three highly unbalanced UCI benchmarks and on one\nreal-world redacted email data set.\n", "date": "2023-07-13", "categories": []}
{"id": "2307.09109", "title": "Mining of Single-Class by Active Learning for Semantic Segmentation", "abstract": "  Several Active Learning (AL) policies require retraining a target model\nseveral times in order to identify the most informative samples and rarely\noffer the option to focus on the acquisition of samples from underrepresented\nclasses. Here the Mining of Single-Class by Active Learning (MiSiCAL) paradigm\nis introduced where an AL policy is constructed through deep reinforcement\nlearning and exploits quantity-accuracy correlations to build datasets on which\nhigh-performance models can be trained with regards to specific classes.\nMiSiCAL is especially helpful in the case of very large batch sizes since it\ndoes not require repeated model training sessions as is common in other AL\nmethods. This is thanks to its ability to exploit fixed representations of the\ncandidate data points. We find that MiSiCAL is able to outperform a random\npolicy on 150 out of 171 COCO10k classes, while the strongest baseline only\noutperforms random on 101 classes.\n", "date": "2023-07-18", "categories": []}
{"id": "2307.07942", "title": "KECOR: Kernel Coding Rate Maximization for Active 3D Object Detection", "abstract": "  Achieving a reliable LiDAR-based object detector in autonomous driving is\nparamount, but its success hinges on obtaining large amounts of precise 3D\nannotations. Active learning (AL) seeks to mitigate the annotation burden\nthrough algorithms that use fewer labels and can attain performance comparable\nto fully supervised learning. Although AL has shown promise, current approaches\nprioritize the selection of unlabeled point clouds with high uncertainty and/or\ndiversity, leading to the selection of more instances for labeling and reduced\ncomputational efficiency. In this paper, we resort to a novel kernel coding\nrate maximization (KECOR) strategy which aims to identify the most informative\npoint clouds to acquire labels through the lens of information theory. Greedy\nsearch is applied to seek desired point clouds that can maximize the minimal\nnumber of bits required to encode the latent features. To determine the\nuniqueness and informativeness of the selected samples from the model\nperspective, we construct a proxy network of the 3D detector head and compute\nthe outer product of Jacobians from all proxy layers to form the empirical\nneural tangent kernel (NTK) matrix. To accommodate both one-stage (i.e.,\nSECOND) and two-stage detectors (i.e., PVRCNN), we further incorporate the\nclassification entropy maximization and well trade-off between detection\nperformance and the total number of bounding boxes selected for annotation.\nExtensive experiments conducted on two 3D benchmarks and a 2D detection dataset\nevidence the superiority and versatility of the proposed approach. Our results\nshow that approximately 44% box-level annotation costs and 26% computational\ntime are reduced compared to the state-of-the-art AL method, without\ncompromising detection performance.\n", "date": "2023-07-16", "categories": []}
{"id": "2307.08414", "title": "Active Learning for Object Detection with Non-Redundant Informative\n  Sampling", "abstract": "  Curating an informative and representative dataset is essential for enhancing\nthe performance of 2D object detectors. We present a novel active learning\nsampling strategy that addresses both the informativeness and diversity of the\nselections. Our strategy integrates uncertainty and diversity-based selection\nprinciples into a joint selection objective by measuring the collective\ninformation score of the selected samples. Specifically, our proposed NORIS\nalgorithm quantifies the impact of training with a sample on the\ninformativeness of other similar samples. By exclusively selecting samples that\nare simultaneously informative and distant from other highly informative\nsamples, we effectively avoid redundancy while maintaining a high level of\ninformativeness. Moreover, instead of utilizing whole image features to\ncalculate distances between samples, we leverage features extracted from\ndetected object regions within images to define object features. This allows us\nto construct a dataset encompassing diverse object types, shapes, and angles.\nExtensive experiments on object detection and image classification tasks\ndemonstrate the effectiveness of our strategy over the state-of-the-art\nbaselines. Specifically, our selection strategy achieves a 20% and 30%\nreduction in labeling costs compared to random selection for PASCAL-VOC and\nKITTI, respectively.\n", "date": "2023-07-17", "categories": []}
{"id": "2207.09874", "title": "Stream-based active learning with linear models", "abstract": "  The proliferation of automated data collection schemes and the advances in\nsensorics are increasing the amount of data we are able to monitor in\nreal-time. However, given the high annotation costs and the time required by\nquality inspections, data is often available in an unlabeled form. This is\nfostering the use of active learning for the development of soft sensors and\npredictive models. In production, instead of performing random inspections to\nobtain product information, labels are collected by evaluating the information\ncontent of the unlabeled data. Several query strategy frameworks for regression\nhave been proposed in the literature but most of the focus has been dedicated\nto the static pool-based scenario. In this work, we propose a new strategy for\nthe stream-based scenario, where instances are sequentially offered to the\nlearner, which must instantaneously decide whether to perform the quality check\nto obtain the label or discard the instance. The approach is inspired by the\noptimal experimental design theory and the iterative aspect of the\ndecision-making process is tackled by setting a threshold on the\ninformativeness of the unlabeled data points. The proposed approach is\nevaluated using numerical simulations and the Tennessee Eastman Process\nsimulator. The results confirm that selecting the examples suggested by the\nproposed algorithm allows for a faster reduction in the prediction error.\n", "date": "2022-07-20", "categories": []}
{"id": "2307.07168", "title": "Adaptive Region Selection for Active Learning in Whole Slide Image\n  Semantic Segmentation", "abstract": "  The process of annotating histological gigapixel-sized whole slide images\n(WSIs) at the pixel level for the purpose of training a supervised segmentation\nmodel is time-consuming. Region-based active learning (AL) involves training\nthe model on a limited number of annotated image regions instead of requesting\nannotations of the entire images. These annotation regions are iteratively\nselected, with the goal of optimizing model performance while minimizing the\nannotated area. The standard method for region selection evaluates the\ninformativeness of all square regions of a specified size and then selects a\nspecific quantity of the most informative regions. We find that the efficiency\nof this method highly depends on the choice of AL step size (i.e., the\ncombination of region size and the number of selected regions per WSI), and a\nsuboptimal AL step size can result in redundant annotation requests or inflated\ncomputation costs. This paper introduces a novel technique for selecting\nannotation regions adaptively, mitigating the reliance on this AL\nhyperparameter. Specifically, we dynamically determine each region by first\nidentifying an informative area and then detecting its optimal bounding box, as\nopposed to selecting regions of a uniform predefined shape and size as in the\nstandard method. We evaluate our method using the task of breast cancer\nmetastases segmentation on the public CAMELYON16 dataset and show that it\nconsistently achieves higher sampling efficiency than the standard method\nacross various AL step sizes. With only 2.6\\% of tissue area annotated, we\nachieve full annotation performance and thereby substantially reduce the costs\nof annotating a WSI dataset. The source code is available at\nhttps://github.com/DeepMicroscopy/AdaptiveRegionSelection.\n", "date": "2023-07-14", "categories": []}
{"id": "2307.07413", "title": "Exploiting Counter-Examples for Active Learning with Partial labels", "abstract": "  This paper studies a new problem, \\emph{active learning with partial labels}\n(ALPL). In this setting, an oracle annotates the query samples with partial\nlabels, relaxing the oracle from the demanding accurate labeling process. To\naddress ALPL, we first build an intuitive baseline that can be seamlessly\nincorporated into existing AL frameworks. Though effective, this baseline is\nstill susceptible to the \\emph{overfitting}, and falls short of the\nrepresentative partial-label-based samples during the query process. Drawing\ninspiration from human inference in cognitive science, where accurate\ninferences can be explicitly derived from \\emph{counter-examples} (CEs), our\nobjective is to leverage this human-like learning pattern to tackle the\n\\emph{overfitting} while enhancing the process of selecting representative\nsamples in ALPL. Specifically, we construct CEs by reversing the partial labels\nfor each instance, and then we propose a simple but effective WorseNet to\ndirectly learn from this complementary pattern. By leveraging the distribution\ngap between WorseNet and the predictor, this adversarial evaluation manner\ncould enhance both the performance of the predictor itself and the sample\nselection process, allowing the predictor to capture more accurate patterns in\nthe data. Experimental results on five real-world datasets and four benchmark\ndatasets show that our proposed method achieves comprehensive improvements over\nten representative AL frameworks, highlighting the superiority of WorseNet. The\nsource code will be available at \\url{https://github.com/Ferenas/APLL}.\n", "date": "2023-07-14", "categories": []}
{"id": "2307.05587", "title": "Active Learning for Video Classification with Frame Level Queries", "abstract": "  Deep learning algorithms have pushed the boundaries of computer vision\nresearch and have depicted commendable performance in a variety of\napplications. However, training a robust deep neural network necessitates a\nlarge amount of labeled training data, acquiring which involves significant\ntime and human effort. This problem is even more serious for an application\nlike video classification, where a human annotator has to watch an entire video\nend-to-end to furnish a label. Active learning algorithms automatically\nidentify the most informative samples from large amounts of unlabeled data;\nthis tremendously reduces the human annotation effort in inducing a machine\nlearning model, as only the few samples that are identified by the algorithm,\nneed to be labeled manually. In this paper, we propose a novel active learning\nframework for video classification, with the goal of further reducing the\nlabeling onus on the human annotators. Our framework identifies a batch of\nexemplar videos, together with a set of informative frames for each video; the\nhuman annotator needs to merely review the frames and provide a label for each\nvideo. This involves much less manual work than watching the complete video to\ncome up with a label. We formulate a criterion based on uncertainty and\ndiversity to identify the informative videos and exploit representative\nsampling techniques to extract a set of exemplar frames from each video. To the\nbest of our knowledge, this is the first research effort to develop an active\nlearning framework for video classification, where the annotators need to\ninspect only a few frames to produce a label, rather than watching the\nend-to-end video.\n", "date": "2023-07-10", "categories": []}
{"id": "2307.05254", "title": "OpenAL: An Efficient Deep Active Learning Framework for Open-Set\n  Pathology Image Classification", "abstract": "  Active learning (AL) is an effective approach to select the most informative\nsamples to label so as to reduce the annotation cost. Existing AL methods\ntypically work under the closed-set assumption, i.e., all classes existing in\nthe unlabeled sample pool need to be classified by the target model. However,\nin some practical clinical tasks, the unlabeled pool may contain not only the\ntarget classes that need to be fine-grainedly classified, but also non-target\nclasses that are irrelevant to the clinical tasks. Existing AL methods cannot\nwork well in this scenario because they tend to select a large number of\nnon-target samples. In this paper, we formulate this scenario as an open-set AL\nproblem and propose an efficient framework, OpenAL, to address the challenge of\nquerying samples from an unlabeled pool with both target class and non-target\nclass samples. Experiments on fine-grained classification of pathology images\nshow that OpenAL can significantly improve the query quality of target class\nsamples and achieve higher performance than current state-of-the-art AL\nmethods. Code is available at https://github.com/miccaiif/OpenAL.\n", "date": "2023-07-11", "categories": []}
{"id": "2302.08981", "title": "Black-Box Batch Active Learning for Regression", "abstract": "  Batch active learning is a popular approach for efficiently training machine\nlearning models on large, initially unlabelled datasets by repeatedly acquiring\nlabels for batches of data points. However, many recent batch active learning\nmethods are white-box approaches and are often limited to differentiable\nparametric models: they score unlabeled points using acquisition functions\nbased on model embeddings or first- and second-order derivatives. In this\npaper, we propose black-box batch active learning for regression tasks as an\nextension of white-box approaches. Crucially, our method only relies on model\npredictions. This approach is compatible with a wide range of machine learning\nmodels, including regular and Bayesian deep learning models and\nnon-differentiable models such as random forests. It is rooted in Bayesian\nprinciples and utilizes recent kernel-based approaches. This allows us to\nextend a wide range of existing state-of-the-art white-box batch active\nlearning methods (BADGE, BAIT, LCMD) to black-box models. We demonstrate the\neffectiveness of our approach through extensive experimental evaluations on\nregression datasets, achieving surprisingly strong performance compared to\nwhite-box approaches for deep learning models.\n", "date": "2023-02-17", "categories": []}
{"id": "2307.03741", "title": "Training Ensembles with Inliers and Outliers for Semi-supervised Active\n  Learning", "abstract": "  Deep active learning in the presence of outlier examples poses a realistic\nyet challenging scenario. Acquiring unlabeled data for annotation requires a\ndelicate balance between avoiding outliers to conserve the annotation budget\nand prioritizing useful inlier examples for effective training. In this work,\nwe present an approach that leverages three highly synergistic components,\nwhich are identified as key ingredients: joint classifier training with inliers\nand outliers, semi-supervised learning through pseudo-labeling, and model\nensembling. Our work demonstrates that ensembling significantly enhances the\naccuracy of pseudo-labeling and improves the quality of data acquisition. By\nenabling semi-supervision through the joint training process, where outliers\nare properly handled, we observe a substantial boost in classifier accuracy\nthrough the use of all available unlabeled examples. Notably, we reveal that\nthe integration of joint training renders explicit outlier detection\nunnecessary; a conventional component for acquisition in prior work. The three\nkey components align seamlessly with numerous existing approaches. Through\nempirical evaluations, we showcase that their combined use leads to a\nperformance increase. Remarkably, despite its simplicity, our proposed approach\noutperforms all other methods in terms of performance. Code:\nhttps://github.com/vladan-stojnic/active-outliers\n", "date": "2023-07-07", "categories": []}
{"id": "2307.00968", "title": "REAL: A Representative Error-Driven Approach for Active Learning", "abstract": "  Given a limited labeling budget, active learning (AL) aims to sample the most\ninformative instances from an unlabeled pool to acquire labels for subsequent\nmodel training. To achieve this, AL typically measures the informativeness of\nunlabeled instances based on uncertainty and diversity. However, it does not\nconsider erroneous instances with their neighborhood error density, which have\ngreat potential to improve the model performance. To address this limitation,\nwe propose $REAL$, a novel approach to select data instances with\n$\\underline{R}$epresentative $\\underline{E}$rrors for $\\underline{A}$ctive\n$\\underline{L}$earning. It identifies minority predictions as \\emph{pseudo\nerrors} within a cluster and allocates an adaptive sampling budget for the\ncluster based on estimated error density. Extensive experiments on five text\nclassification datasets demonstrate that $REAL$ consistently outperforms all\nbest-performing baselines regarding accuracy and F1-macro scores across a wide\nrange of hyperparameter settings. Our analysis also shows that $REAL$ selects\nthe most representative pseudo errors that match the distribution of\nground-truth errors along the decision boundary. Our code is publicly available\nat https://github.com/withchencheng/ECML_PKDD_23_Real.\n", "date": "2023-07-03", "categories": []}
{"id": "2307.02641", "title": "Active Class Selection for Few-Shot Class-Incremental Learning", "abstract": "  For real-world applications, robots will need to continually learn in their\nenvironments through limited interactions with their users. Toward this,\nprevious works in few-shot class incremental learning (FSCIL) and active class\nselection (ACS) have achieved promising results but were tested in constrained\nsetups. Therefore, in this paper, we combine ideas from FSCIL and ACS to\ndevelop a novel framework that can allow an autonomous agent to continually\nlearn new objects by asking its users to label only a few of the most\ninformative objects in the environment. To this end, we build on a\nstate-of-the-art (SOTA) FSCIL model and extend it with techniques from ACS\nliterature. We term this model Few-shot Incremental Active class SeleCtiOn\n(FIASco). We further integrate a potential field-based navigation technique\nwith our model to develop a complete framework that can allow an agent to\nprocess and reason on its sensory data through the FIASco model, navigate\ntowards the most informative object in the environment, gather data about the\nobject through its sensors and incrementally update the FIASco model.\nExperimental results on a simulated agent and a real robot show the\nsignificance of our approach for long-term real-world robotics applications.\n", "date": "2023-07-05", "categories": []}
{"id": "2304.06808", "title": "Active Cost-aware Labeling of Streaming Data", "abstract": "  We study actively labeling streaming data, where an active learner is faced\nwith a stream of data points and must carefully choose which of these points to\nlabel via an expensive experiment. Such problems frequently arise in\napplications such as healthcare and astronomy. We first study a setting when\nthe data's inputs belong to one of $K$ discrete distributions and formalize\nthis problem via a loss that captures the labeling cost and the prediction\nerror. When the labeling cost is $B$, our algorithm, which chooses to label a\npoint if the uncertainty is larger than a time and cost dependent threshold,\nachieves a worst-case upper bound of $\\widetilde{O}(B^{\\frac{1}{3}}\nK^{\\frac{1}{3}} T^{\\frac{2}{3}})$ on the loss after $T$ rounds. We also provide\na more nuanced upper bound which demonstrates that the algorithm can adapt to\nthe arrival pattern, and achieves better performance when the arrival pattern\nis more favorable. We complement both upper bounds with matching lower bounds.\nWe next study this problem when the inputs belong to a continuous domain and\nthe output of the experiment is a smooth function with bounded RKHS norm. After\n$T$ rounds in $d$ dimensions, we show that the loss is bounded by\n$\\widetilde{O}(B^{\\frac{1}{d+3}} T^{\\frac{d+2}{d+3}})$ in an RKHS with a\nsquared exponential kernel and by $\\widetilde{O}(B^{\\frac{1}{2d+3}}\nT^{\\frac{2d+2}{2d+3}})$ in an RKHS with a Mat\\'ern kernel. Our empirical\nevaluation demonstrates that our method outperforms other baselines in several\nsynthetic experiments and two real experiments in medicine and astronomy.\n", "date": "2023-04-13", "categories": []}
{"id": "2307.01582", "title": "IAdet: Simplest human-in-the-loop object detection", "abstract": "  This work proposes a strategy for training models while annotating data named\nIntelligent Annotation (IA). IA involves three modules: (1) assisted data\nannotation, (2) background model training, and (3) active selection of the next\ndatapoints. Under this framework, we open-source the IAdet tool, which is\nspecific for single-class object detection. Additionally, we devise a method\nfor automatically evaluating such a human-in-the-loop system. For the PASCAL\nVOC dataset, the IAdet tool reduces the database annotation time by $25\\%$\nwhile providing a trained model for free. These results are obtained for a\ndeliberately very simple IAdet design. As a consequence, IAdet is susceptible\nto multiple easy improvements, paving the way for powerful human-in-the-loop\nobject detection systems.\n", "date": "2023-07-04", "categories": []}
{"id": "2211.05039", "title": "Active Acquisition for Multimodal Temporal Data: A Challenging\n  Decision-Making Task", "abstract": "  We introduce a challenging decision-making task that we call active\nacquisition for multimodal temporal data (A2MT). In many real-world scenarios,\ninput features are not readily available at test time and must instead be\nacquired at significant cost. With A2MT, we aim to learn agents that actively\nselect which modalities of an input to acquire, trading off acquisition cost\nand predictive performance. A2MT extends a previous task called active feature\nacquisition to temporal decision making about high-dimensional inputs. We\npropose a method based on the Perceiver IO architecture to address A2MT in\npractice. Our agents are able to solve a novel synthetic scenario requiring\npractically relevant cross-modal reasoning skills. On two large-scale,\nreal-world datasets, Kinetics-700 and AudioSet, our agents successfully learn\ncost-reactive acquisition behavior. However, an ablation reveals they are\nunable to learn adaptive acquisition strategies, emphasizing the difficulty of\nthe task even for state-of-the-art models. Applications of A2MT may be\nimpactful in domains like medicine, robotics, or finance, where modalities\ndiffer in acquisition cost and informativeness.\n", "date": "2022-11-09", "categories": []}
{"id": "2306.16918", "title": "PCDAL: A Perturbation Consistency-Driven Active Learning Approach for\n  Medical Image Segmentation and Classification", "abstract": "  In recent years, deep learning has become a breakthrough technique in\nassisting medical image diagnosis. Supervised learning using convolutional\nneural networks (CNN) provides state-of-the-art performance and has served as a\nbenchmark for various medical image segmentation and classification. However,\nsupervised learning deeply relies on large-scale annotated data, which is\nexpensive, time-consuming, and even impractical to acquire in medical imaging\napplications. Active Learning (AL) methods have been widely applied in natural\nimage classification tasks to reduce annotation costs by selecting more\nvaluable examples from the unlabeled data pool. However, their application in\nmedical image segmentation tasks is limited, and there is currently no\neffective and universal AL-based method specifically designed for 3D medical\nimage segmentation. To address this limitation, we propose an AL-based method\nthat can be simultaneously applied to 2D medical image classification,\nsegmentation, and 3D medical image segmentation tasks. We extensively validated\nour proposed active learning method on three publicly available and challenging\nmedical image datasets, Kvasir Dataset, COVID-19 Infection Segmentation\nDataset, and BraTS2019 Dataset. The experimental results demonstrate that our\nPCDAL can achieve significantly improved performance with fewer annotations in\n2D classification and segmentation and 3D segmentation tasks. The codes of this\nstudy are available at https://github.com/ortonwang/PCDAL.\n", "date": "2023-06-29", "categories": []}
{"id": "2306.15766", "title": "Large Language Models as Annotators: Enhancing Generalization of NLP\n  Models at Minimal Cost", "abstract": "  State-of-the-art supervised NLP models achieve high accuracy but are also\nsusceptible to failures on inputs from low-data regimes, such as domains that\nare not represented in training data. As an approximation to collecting\nground-truth labels for the specific domain, we study the use of large language\nmodels (LLMs) for annotating inputs and improving the generalization of NLP\nmodels. Specifically, given a budget for LLM annotations, we present an\nalgorithm for sampling the most informative inputs to annotate and retrain the\nNLP model. We find that popular active learning strategies such as\nuncertainty-based sampling do not work well. Instead, we propose a sampling\nstrategy based on the difference in prediction scores between the base model\nand the finetuned NLP model, utilizing the fact that most NLP models are\nfinetuned from a base model. Experiments with classification (semantic\nsimilarity) and ranking (semantic search) tasks show that our sampling strategy\nleads to significant gains in accuracy for both the training and target\ndomains.\n", "date": "2023-06-27", "categories": []}
{"id": "2306.15058", "title": "BatchGFN: Generative Flow Networks for Batch Active Learning", "abstract": "  We introduce BatchGFN -- a novel approach for pool-based active learning that\nuses generative flow networks to sample sets of data points proportional to a\nbatch reward. With an appropriate reward function to quantify the utility of\nacquiring a batch, such as the joint mutual information between the batch and\nthe model parameters, BatchGFN is able to construct highly informative batches\nfor active learning in a principled way. We show our approach enables sampling\nnear-optimal utility batches at inference time with a single forward pass per\npoint in the batch in toy regression problems. This alleviates the\ncomputational complexity of batch-aware algorithms and removes the need for\ngreedy approximations to find maximizers for the batch reward. We also present\nearly results for amortizing training across acquisition steps, which will\nenable scaling to real-world tasks.\n", "date": "2023-06-26", "categories": []}
{"id": "2306.11605", "title": "Annotation Cost Efficient Active Learning for Content Based Image\n  Retrieval", "abstract": "  Deep metric learning (DML) based methods have been found very effective for\ncontent-based image retrieval (CBIR) in remote sensing (RS). For accurately\nlearning the model parameters of deep neural networks, most of the DML methods\nrequire a high number of annotated training images, which can be costly to\ngather. To address this problem, in this paper we present an annotation cost\nefficient active learning (AL) method (denoted as ANNEAL). The proposed method\naims to iteratively enrich the training set by annotating the most informative\nimage pairs as similar or dissimilar, while accurately modelling a deep metric\nspace. This is achieved by two consecutive steps. In the first step the\npairwise image similarity is modelled based on the available training set.\nThen, in the second step the most uncertain and diverse (i.e., informative)\nimage pairs are selected to be annotated. Unlike the existing AL methods for\nCBIR, at each AL iteration of ANNEAL a human expert is asked to annotate the\nmost informative image pairs as similar/dissimilar. This significantly reduces\nthe annotation cost compared to annotating images with land-use/land cover\nclass labels. Experimental results show the effectiveness of our method. The\ncode of ANNEAL is publicly available at https://git.tu-berlin.de/rsim/ANNEAL.\n", "date": "2023-06-20", "categories": []}
{"id": "2306.13923", "title": "Active Data Acquisition in Autonomous Driving Simulation", "abstract": "  Autonomous driving algorithms rely heavily on learning-based models, which\nrequire large datasets for training. However, there is often a large amount of\nredundant information in these datasets, while collecting and processing these\ndatasets can be time-consuming and expensive. To address this issue, this paper\nproposes the concept of an active data-collecting strategy. For high-quality\ndata, increasing the collection density can improve the overall quality of the\ndataset, ultimately achieving similar or even better results than the original\ndataset with lower labeling costs and smaller dataset sizes. In this paper, we\ndesign experiments to verify the quality of the collected dataset and to\ndemonstrate this strategy can significantly reduce labeling costs and dataset\nsize while improving the overall quality of the dataset, leading to better\nperformance of autonomous driving systems. The source code implementing the\nproposed approach is publicly available on\nhttps://github.com/Th1nkMore/carla_dataset_tools.\n", "date": "2023-06-24", "categories": []}
{"id": "2306.14347", "title": "Fast Classification with Sequential Feature Selection in Test Phase", "abstract": "  This paper introduces a novel approach to active feature acquisition for\nclassification, which is the task of sequentially selecting the most\ninformative subset of features to achieve optimal prediction performance during\ntesting while minimizing cost. The proposed approach involves a new lazy model\nthat is significantly faster and more efficient compared to existing methods,\nwhile still producing comparable accuracy results. During the test phase, the\nproposed approach utilizes Fisher scores for feature ranking to identify the\nmost important feature at each step. In the next step the training dataset is\nfiltered based on the observed value of the selected feature and then we\ncontinue this process to reach to acceptable accuracy or limit of the budget\nfor feature acquisition. The performance of the proposed approach was evaluated\non synthetic and real datasets, including our new synthetic dataset, CUBE\ndataset and also real dataset Forest. The experimental results demonstrate that\nour approach achieves competitive accuracy results compared to existing\nmethods, while significantly outperforming them in terms of speed. The source\ncode of the algorithm is released at github with this link:\nhttps://github.com/alimirzaei/FCwSFS.\n", "date": "2023-06-25", "categories": []}
{"id": "2306.06908", "title": "Active Learning Guided Fine-Tuning for enhancing Self-Supervised Based\n  Multi-Label Classification of Remote Sensing Images", "abstract": "  In recent years, deep neural networks (DNNs) have been found very successful\nfor multi-label classification (MLC) of remote sensing (RS) images.\nSelf-supervised pre-training combined with fine-tuning on a randomly selected\nsmall training set has become a popular approach to minimize annotation efforts\nof data-demanding DNNs. However, fine-tuning on a small and biased training set\nmay limit model performance. To address this issue, we investigate the\neffectiveness of the joint use of self-supervised pre-training with active\nlearning (AL). The considered AL strategy aims at guiding the MLC fine-tuning\nof a self-supervised model by selecting informative training samples to\nannotate in an iterative manner. Experimental results show the effectiveness of\napplying AL-guided fine-tuning (particularly for the case where strong\nclass-imbalance is present in MLC problems) compared to the application of\nfine-tuning using a randomly constructed small training set.\n", "date": "2023-06-12", "categories": []}
{"id": "2306.12376", "title": "M-VAAL: Multimodal Variational Adversarial Active Learning for\n  Downstream Medical Image Analysis Tasks", "abstract": "  Acquiring properly annotated data is expensive in the medical field as it\nrequires experts, time-consuming protocols, and rigorous validation. Active\nlearning attempts to minimize the need for large annotated samples by actively\nsampling the most informative examples for annotation. These examples\ncontribute significantly to improving the performance of supervised machine\nlearning models, and thus, active learning can play an essential role in\nselecting the most appropriate information in deep learning-based diagnosis,\nclinical assessments, and treatment planning. Although some existing works have\nproposed methods for sampling the best examples for annotation in medical image\nanalysis, they are not task-agnostic and do not use multimodal auxiliary\ninformation in the sampler, which has the potential to increase robustness.\nTherefore, in this work, we propose a Multimodal Variational Adversarial Active\nLearning (M-VAAL) method that uses auxiliary information from additional\nmodalities to enhance the active sampling. We applied our method to two\ndatasets: i) brain tumor segmentation and multi-label classification using the\nBraTS2018 dataset, and ii) chest X-ray image classification using the\nCOVID-QU-Ex dataset. Our results show a promising direction toward\ndata-efficient learning under limited annotations.\n", "date": "2023-06-21", "categories": []}
{"id": "2306.12398", "title": "Multi-Task Consistency for Active Learning", "abstract": "  Learning-based solutions for vision tasks require a large amount of labeled\ntraining data to ensure their performance and reliability. In single-task\nvision-based settings, inconsistency-based active learning has proven to be\neffective in selecting informative samples for annotation. However, there is a\nlack of research exploiting the inconsistency between multiple tasks in\nmulti-task networks. To address this gap, we propose a novel multi-task active\nlearning strategy for two coupled vision tasks: object detection and semantic\nsegmentation. Our approach leverages the inconsistency between them to identify\ninformative samples across both tasks. We propose three constraints that\nspecify how the tasks are coupled and introduce a method for determining the\npixels belonging to the object detected by a bounding box, to later quantify\nthe constraints as inconsistency scores. To evaluate the effectiveness of our\napproach, we establish multiple baselines for multi-task active learning and\nintroduce a new metric, mean Detection Segmentation Quality (mDSQ), tailored\nfor the multi-task active learning comparison that addresses the performance of\nboth tasks. We conduct extensive experiments on the nuImages and A9 datasets,\ndemonstrating that our approach outperforms existing state-of-the-art methods\nby up to 3.4% mDSQ on nuImages. Our approach achieves 95% of the fully-trained\nperformance using only 67% of the available data, corresponding to 20% fewer\nlabels compared to random selection and 5% fewer labels compared to\nstate-of-the-art selection strategy. Our code will be made publicly available\nafter the review process.\n", "date": "2023-06-21", "categories": []}
{"id": "2303.04912", "title": "Embodied Active Learning of Relational State Abstractions for Bilevel\n  Planning", "abstract": "  State abstraction is an effective technique for planning in robotics\nenvironments with continuous states and actions, long task horizons, and sparse\nfeedback. In object-oriented environments, predicates are a particularly useful\nform of state abstraction because of their compatibility with symbolic planners\nand their capacity for relational generalization. However, to plan with\npredicates, the agent must be able to interpret them in continuous environment\nstates (i.e., ground the symbols). Manually programming predicate\ninterpretations can be difficult, so we would instead like to learn them from\ndata. We propose an embodied active learning paradigm where the agent learns\npredicate interpretations through online interaction with an expert. For\nexample, after taking actions in a block stacking environment, the agent may\nask the expert: \"Is On(block1, block2) true?\" From this experience, the agent\nlearns to plan: it learns neural predicate interpretations, symbolic planning\noperators, and neural samplers that can be used for bilevel planning. During\nexploration, the agent plans to learn: it uses its current models to select\nactions towards generating informative expert queries. We learn predicate\ninterpretations as ensembles of neural networks and use their entropy to\nmeasure the informativeness of potential queries. We evaluate this approach in\nthree robotic environments and find that it consistently outperforms six\nbaselines while exhibiting sample efficiency in two key metrics: number of\nenvironment interactions, and number of queries to the expert. Code:\nhttps://tinyurl.com/active-predicates\n", "date": "2023-03-08", "categories": []}
{"id": "2304.04389", "title": "Deep Active Alignment of Knowledge Graph Entities and Schemata", "abstract": "  Knowledge graphs (KGs) store rich facts about the real world. In this paper,\nwe study KG alignment, which aims to find alignment between not only entities\nbut also relations and classes in different KGs. Alignment at the entity level\ncan cross-fertilize alignment at the schema level. We propose a new KG\nalignment approach, called DAAKG, based on deep learning and active learning.\nWith deep learning, it learns the embeddings of entities, relations and\nclasses, and jointly aligns them in a semi-supervised manner. With active\nlearning, it estimates how likely an entity, relation or class pair can be\ninferred, and selects the best batch for human labeling. We design two\napproximation algorithms for efficient solution to batch selection. Our\nexperiments on benchmark datasets show the superior accuracy and generalization\nof DAAKG and validate the effectiveness of all its modules.\n", "date": "2023-04-10", "categories": []}
{"id": "2305.13402", "title": "Error-Tolerant Exact Query Learning of Finite Set Partitions with\n  Same-Cluster Oracle", "abstract": "  This paper initiates the study of active learning for exact recovery of\npartitions exclusively through access to a same-cluster oracle in the presence\nof bounded adversarial error. We first highlight a novel connection between\nlearning partitions and correlation clustering. Then we use this connection to\nbuild a R\\'enyi-Ulam style analytical framework for this problem, and prove\nupper and lower bounds on its worst-case query complexity. Further, we bound\nthe expected performance of a relevant randomized algorithm. Finally, we study\nthe relationship between adaptivity and query complexity for this problem and\nrelated variants.\n", "date": "2023-05-22", "categories": []}
{"id": "2306.10087", "title": "ActiveGLAE: A Benchmark for Deep Active Learning with Transformers", "abstract": "  Deep active learning (DAL) seeks to reduce annotation costs by enabling the\nmodel to actively query instance annotations from which it expects to learn the\nmost. Despite extensive research, there is currently no standardized evaluation\nprotocol for transformer-based language models in the field of DAL. Diverse\nexperimental settings lead to difficulties in comparing research and deriving\nrecommendations for practitioners. To tackle this challenge, we propose the\nActiveGLAE benchmark, a comprehensive collection of data sets and evaluation\nguidelines for assessing DAL. Our benchmark aims to facilitate and streamline\nthe evaluation process of novel DAL strategies. Additionally, we provide an\nextensive overview of current practice in DAL with transformer-based language\nmodels. We identify three key challenges - data set selection, model training,\nand DAL settings - that pose difficulties in comparing query strategies. We\nestablish baseline results through an extensive set of experiments as a\nreference point for evaluating future work. Based on our findings, we provide\nguidelines for researchers and practitioners.\n", "date": "2023-06-16", "categories": []}
{"id": "2306.10700", "title": "Perturbation-Based Two-Stage Multi-Domain Active Learning", "abstract": "  In multi-domain learning (MDL) scenarios, high labeling effort is required\ndue to the complexity of collecting data from various domains. Active Learning\n(AL) presents an encouraging solution to this issue by annotating a smaller\nnumber of highly informative instances, thereby reducing the labeling effort.\nPrevious research has relied on conventional AL strategies for MDL scenarios,\nwhich underutilize the domain-shared information of each instance during the\nselection procedure. To mitigate this issue, we propose a novel\nperturbation-based two-stage multi-domain active learning (P2S-MDAL) method\nincorporated into the well-regarded ASP-MTL model. Specifically, P2S-MDAL\ninvolves allocating budgets for domains and establishing regions for diversity\nselection, which are further used to select the most cross-domain influential\nsamples in each region. A perturbation metric has been introduced to evaluate\nthe robustness of the shared feature extractor of the model, facilitating the\nidentification of potentially cross-domain influential samples. Experiments are\nconducted on three real-world datasets, encompassing both texts and images. The\nsuperior performance over conventional AL strategies shows the effectiveness of\nthe proposed strategy. Additionally, an ablation study has been carried out to\ndemonstrate the validity of each component. Finally, we outline several\nintriguing potential directions for future MDAL research, thus catalyzing the\nfield's advancement.\n", "date": "2023-06-19", "categories": []}
{"id": "2306.11056", "title": "Taming Small-sample Bias in Low-budget Active Learning", "abstract": "  Active learning (AL) aims to minimize the annotation cost by only querying a\nfew informative examples for each model training stage. However, training a\nmodel on a few queried examples suffers from the small-sample bias. In this\npaper, we address this small-sample bias issue in low-budget AL by exploring a\nregularizer called Firth bias reduction, which can provably reduce the bias\nduring the model training process but might hinder learning if its coefficient\nis not adaptive to the learning progress. Instead of tuning the coefficient for\neach query round, which is sensitive and time-consuming, we propose the\ncurriculum Firth bias reduction (CHAIN) that can automatically adjust the\ncoefficient to be adaptive to the training process. Under both deep learning\nand linear model settings, experiments on three benchmark datasets with several\nwidely used query strategies and hyperparameter searching methods show that\nCHAIN can be used to build more efficient AL and can substantially improve the\nprogress made by each active learning query.\n", "date": "2023-06-19", "categories": []}
{"id": "2302.04332", "title": "Continuous Learning for Android Malware Detection", "abstract": "  Machine learning methods can detect Android malware with very high accuracy.\nHowever, these classifiers have an Achilles heel, concept drift: they rapidly\nbecome out of date and ineffective, due to the evolution of malware apps and\nbenign apps. Our research finds that, after training an Android malware\nclassifier on one year's worth of data, the F1 score quickly dropped from 0.99\nto 0.76 after 6 months of deployment on new test samples.\n  In this paper, we propose new methods to combat the concept drift problem of\nAndroid malware classifiers. Since machine learning technique needs to be\ncontinuously deployed, we use active learning: we select new samples for\nanalysts to label, and then add the labeled samples to the training set to\nretrain the classifier. Our key idea is, similarity-based uncertainty is more\nrobust against concept drift. Therefore, we combine contrastive learning with\nactive learning. We propose a new hierarchical contrastive learning scheme, and\na new sample selection technique to continuously train the Android malware\nclassifier. Our evaluation shows that this leads to significant improvements,\ncompared to previously published methods for active learning. Our approach\nreduces the false negative rate from 14% (for the best baseline) to 9%, while\nalso reducing the false positive rate (from 0.86% to 0.48%). Also, our approach\nmaintains more consistent performance across a seven-year time period than past\nmethods.\n", "date": "2023-02-08", "categories": []}
{"id": "2306.08001", "title": "A Markovian Formalism for Active Querying", "abstract": "  Active learning algorithms have been an integral part of recent advances in\nartificial intelligence. However, the research in the field is widely varying\nand lacks an overall organizing leans. We outline a Markovian formalism for the\nfield of active learning and survey the literature to demonstrate the\norganizing capability of our proposed formalism. Our formalism takes a\npartially observable Markovian system approach to the active learning process\nas a whole. We specifically outline how querying, dataset augmentation, reward\nupdates, and other aspects of active learning can be viewed as a transition\nbetween meta-states in a Markovian system, and give direction into how other\naspects of active learning can fit into our formalism.\n", "date": "2023-06-13", "categories": []}
{"id": "2211.10934", "title": "Active Exploration based on Information Gain by Particle Filter for\n  Efficient Spatial Concept Formation", "abstract": "  Autonomous robots need to learn the categories of various places by exploring\ntheir environments and interacting with users. However, preparing training\ndatasets with linguistic instructions from users is time-consuming and\nlabor-intensive. Moreover, effective exploration is essential for appropriate\nconcept formation and rapid environmental coverage. To address this issue, we\npropose an active inference method, referred to as spatial concept formation\nwith information gain-based active exploration (SpCoAE) that combines\nsequential Bayesian inference using particle filters and information gain-based\ndestination determination in a probabilistic generative model. This study\ninterprets the robot's action as a selection of destinations to ask the user,\n`What kind of place is this?' in the context of active inference. This study\nprovides insights into the technical aspects of the proposed method, including\nactive perception and exploration by the robot, and how the method can enable\nmobile robots to learn spatial concepts through active exploration. Our\nexperiment demonstrated the effectiveness of the SpCoAE in efficiently\ndetermining a destination for learning appropriate spatial concepts in home\nenvironments.\n", "date": "2022-11-20", "categories": []}
{"id": "2306.06174", "title": "Active-Learning-Driven Surrogate Modeling for Efficient Simulation of\n  Parametric Nonlinear Systems", "abstract": "  When repeated evaluations for varying parameter configurations of a\nhigh-fidelity physical model are required, surrogate modeling techniques based\non model order reduction are desired. In absence of the governing equations\ndescribing the dynamics, we need to construct the parametric reduced-order\nsurrogate model in a non-intrusive fashion. In this setting, the usual\nresidual-based error estimate for optimal parameter sampling associated with\nthe reduced basis method is not directly available. Our work provides a\nnon-intrusive optimality criterion to efficiently populate the parameter\nsnapshots, thereby, enabling us to effectively construct a parametric surrogate\nmodel. We consider separate parameter-specific proper orthogonal decomposition\n(POD) subspaces and propose an active-learning-driven surrogate model using\nkernel-based shallow neural networks, abbreviated as ActLearn-POD-KSNN\nsurrogate model. To demonstrate the validity of our proposed ideas, we present\nnumerical experiments using two physical models, namely Burgers' equation and\nshallow water equations. Both the models have mixed -- convective and diffusive\n-- effects within their respective parameter domains, with each of them\ndominating in certain regions. The proposed ActLearn-POD-KSNN surrogate model\nefficiently predicts the solution at new parameter locations, even for a\nsetting with multiple interacting shock profiles.\n", "date": "2023-06-09", "categories": []}
{"id": "2306.05331", "title": "Actively learning a Bayesian matrix fusion model with deep side\n  information", "abstract": "  High-dimensional deep neural network representations of images and concepts\ncan be aligned to predict human annotations of diverse stimuli. However, such\nalignment requires the costly collection of behavioral responses, such that, in\npractice, the deep-feature spaces are only ever sparsely sampled. Here, we\npropose an active learning approach to adaptively sampling experimental stimuli\nto efficiently learn a Bayesian matrix factorization model with deep side\ninformation. We observe a significant efficiency gain over a passive baseline.\nFurthermore, with a sequential batched sampling strategy, the algorithm is\napplicable not only to small datasets collected from traditional laboratory\nexperiments but also to settings where large-scale crowdsourced data collection\nis needed to accurately align the high-dimensional deep feature representations\nderived from pre-trained networks.\n", "date": "2023-06-08", "categories": []}
{"id": "2303.02535", "title": "Streaming Active Learning with Deep Neural Networks", "abstract": "  Active learning is perhaps most naturally posed as an online learning\nproblem. However, prior active learning approaches with deep neural networks\nassume offline access to the entire dataset ahead of time. This paper proposes\nVeSSAL, a new algorithm for batch active learning with deep neural networks in\nstreaming settings, which samples groups of points to query for labels at the\nmoment they are encountered. Our approach trades off between uncertainty and\ndiversity of queried samples to match a desired query rate without requiring\nany hand-tuned hyperparameters. Altogether, we expand the applicability of deep\nneural networks to realistic active learning scenarios, such as applications\nrelevant to HCI and large, fractured datasets.\n", "date": "2023-03-05", "categories": []}
{"id": "2306.04099", "title": "NTKCPL: Active Learning on Top of Self-Supervised Model by Estimating\n  True Coverage", "abstract": "  High annotation cost for training machine learning classifiers has driven\nextensive research in active learning and self-supervised learning. Recent\nresearch has shown that in the context of supervised learning different active\nlearning strategies need to be applied at various stages of the training\nprocess to ensure improved performance over the random baseline. We refer to\nthe point where the number of available annotations changes the suitable active\nlearning strategy as the phase transition point. In this paper, we establish\nthat when combining active learning with self-supervised models to achieve\nimproved performance, the phase transition point occurs earlier. It becomes\nchallenging to determine which strategy should be used for previously unseen\ndatasets. We argue that existing active learning algorithms are heavily\ninfluenced by the phase transition because the empirical risk over the entire\nactive learning pool estimated by these algorithms is inaccurate and influenced\nby the number of labeled samples. To address this issue, we propose a novel\nactive learning strategy, neural tangent kernel clustering-pseudo-labels\n(NTKCPL). It estimates empirical risk based on pseudo-labels and the model\nprediction with NTK approximation. We analyze the factors affecting this\napproximation error and design a pseudo-label clustering generation method to\nreduce the approximation error. We validate our method on five datasets,\nempirically demonstrating that it outperforms the baseline methods in most\ncases and is valid over a wider range of training budgets.\n", "date": "2023-06-07", "categories": []}
{"id": "2306.04454", "title": "Training-Free Neural Active Learning with Initialization-Robustness\n  Guarantees", "abstract": "  Existing neural active learning algorithms have aimed to optimize the\npredictive performance of neural networks (NNs) by selecting data for\nlabelling. However, other than a good predictive performance, being robust\nagainst random parameter initializations is also a crucial requirement in\nsafety-critical applications. To this end, we introduce our expected variance\nwith Gaussian processes (EV-GP) criterion for neural active learning, which is\ntheoretically guaranteed to select data points which lead to trained NNs with\nboth (a) good predictive performances and (b) initialization robustness.\nImportantly, our EV-GP criterion is training-free, i.e., it does not require\nany training of the NN during data selection, which makes it computationally\nefficient. We empirically demonstrate that our EV-GP criterion is highly\ncorrelated with both initialization robustness and generalization performance,\nand show that it consistently outperforms baseline methods in terms of both\ndesiderata, especially in situations with limited initial data or large batch\nsizes.\n", "date": "2023-06-07", "categories": []}
{"id": "2306.03356", "title": "Query Complexity of Active Learning for Function Family With Nearly\n  Orthogonal Basis", "abstract": "  Many machine learning algorithms require large numbers of labeled data to\ndeliver state-of-the-art results. In applications such as medical diagnosis and\nfraud detection, though there is an abundance of unlabeled data, it is costly\nto label the data by experts, experiments, or simulations. Active learning\nalgorithms aim to reduce the number of required labeled data points while\npreserving performance. For many convex optimization problems such as linear\nregression and $p$-norm regression, there are theoretical bounds on the number\nof required labels to achieve a certain accuracy. We call this the query\ncomplexity of active learning. However, today's active learning algorithms\nrequire the underlying learned function to have an orthogonal basis. For\nexample, when applying active learning to linear regression, the requirement is\nthe target function is a linear composition of a set of orthogonal linear\nfunctions, and active learning can find the coefficients of these linear\nfunctions. We present a theoretical result to show that active learning does\nnot need an orthogonal basis but rather only requires a nearly orthogonal\nbasis. We provide the corresponding theoretical proofs for the function family\nof nearly orthogonal basis, and its applications associated with the\nalgorithmically efficient active learning framework.\n", "date": "2023-06-06", "categories": []}
{"id": "2305.04392", "title": "Disentangled Multi-Fidelity Deep Bayesian Active Learning", "abstract": "  To balance quality and cost, various domain areas of science and engineering\nrun simulations at multiple levels of sophistication. Multi-fidelity active\nlearning aims to learn a direct mapping from input parameters to simulation\noutputs at the highest fidelity by actively acquiring data from multiple\nfidelity levels. However, existing approaches based on Gaussian processes are\nhardly scalable to high-dimensional data. Deep learning-based methods often\nimpose a hierarchical structure in hidden representations, which only supports\npassing information from low-fidelity to high-fidelity. These approaches can\nlead to the undesirable propagation of errors from low-fidelity representations\nto high-fidelity ones. We propose a novel framework called Disentangled\nMulti-fidelity Deep Bayesian Active Learning (D-MFDAL), which learns the\nsurrogate models conditioned on the distribution of functions at multiple\nfidelities. On benchmark tasks of learning deep surrogates of partial\ndifferential equations including heat equation, Poisson's equation and fluid\nsimulations, our approach significantly outperforms state-of-the-art in\nprediction accuracy and sample efficiency.\n", "date": "2023-05-07", "categories": []}
{"id": "2306.01922", "title": "Agnostic Multi-Group Active Learning", "abstract": "  Inspired by the problem of improving classification accuracy on rare or hard\nsubsets of a population, there has been recent interest in models of learning\nwhere the goal is to generalize to a collection of distributions, each\nrepresenting a ``group''. We consider a variant of this problem from the\nperspective of active learning, where the learner is endowed with the power to\ndecide which examples are labeled from each distribution in the collection, and\nthe goal is to minimize the number of label queries while maintaining\nPAC-learning guarantees. Our main challenge is that standard active learning\ntechniques such as disagreement-based active learning do not directly apply to\nthe multi-group learning objective. We modify existing algorithms to provide a\nconsistent active learning algorithm for an agnostic formulation of multi-group\nlearning, which given a collection of $G$ distributions and a hypothesis class\n$\\mathcal{H}$ with VC-dimension $d$, outputs an $\\epsilon$-optimal hypothesis\nusing $\\tilde{O}\\left( (\\nu^2/\\epsilon^2+1) G d \\theta_{\\mathcal{G}}^2\n\\log^2(1/\\epsilon) + G\\log(1/\\epsilon)/\\epsilon^2 \\right)$ label queries, where\n$\\theta_{\\mathcal{G}}$ is the worst-case disagreement coefficient over the\ncollection. Roughly speaking, this guarantee improves upon the label complexity\nof standard multi-group learning in regimes where disagreement-based active\nlearning algorithms may be expected to succeed, and the number of groups is not\ntoo large. We also consider the special case where each distribution in the\ncollection is individually realizable with respect to $\\mathcal{H}$, and\ndemonstrate $\\tilde{O}\\left( G d \\theta_{\\mathcal{G}} \\log(1/\\epsilon) \\right)$\nlabel queries are sufficient for learning in this case. We further give an\napproximation result for the full agnostic case inspired by the group\nrealizable strategy.\n", "date": "2023-06-02", "categories": []}
{"id": "2306.02628", "title": "Active Ranking of Experts Based on their Performances in Many Tasks", "abstract": "  We consider the problem of ranking n experts based on their performances on d\ntasks. We make a monotonicity assumption stating that for each pair of experts,\none outperforms the other on all tasks. We consider the sequential setting\nwhere in each round, the learner has access to noisy evaluations of actively\nchosen pair of expert-task, given the information available up to the actual\nround. Given a confidence parameter $\\delta$ $\\in$ (0, 1), we provide\nstrategies allowing to recover the correct ranking of experts and develop a\nbound on the total number of queries made by our algorithm that hold with\nprobability at least 1 -- $\\delta$. We show that our strategy is adaptive to\nthe complexity of the problem (our bounds are instance dependent), and develop\nmatching lower bounds up to a poly-logarithmic factor. Finally, we adapt our\nstrategy to the relaxed problem of best expert identification and provide\nnumerical simulation consistent with our theoretical results.\n", "date": "2023-06-05", "categories": []}
{"id": "2306.02808", "title": "Deep Active Learning with Structured Neural Depth Search", "abstract": "  Previous work optimizes traditional active learning (AL) processes with\nincremental neural network architecture search (Active-iNAS) based on data\ncomplexity change, which improves the accuracy and learning efficiency.\nHowever, Active-iNAS trains several models and selects the model with the best\ngeneralization performance for querying the subsequent samples after each\nactive learning cycle. The independent training processes lead to an\ninsufferable computational budget, which is significantly inefficient and\nlimits search flexibility and final performance. To address this issue, we\npropose a novel active strategy with the method called structured variational\ninference (SVI) or structured neural depth search (SNDS) whereby we could use\nthe gradient descent method in neural network depth search during AL processes.\nAt the same time, we theoretically demonstrate that the current VI-based\nmethods based on the mean-field assumption could lead to poor performance. We\napply our strategy using three querying techniques and three datasets and show\nthat our strategy outperforms current methods.\n", "date": "2023-06-05", "categories": []}
{"id": "2212.00772", "title": "Margin-based sampling in high dimensions: When being active is less\n  efficient than staying passive", "abstract": "  It is widely believed that given the same labeling budget, active learning\n(AL) algorithms like margin-based active learning achieve better predictive\nperformance than passive learning (PL), albeit at a higher computational cost.\nRecent empirical evidence suggests that this added cost might be in vain, as\nmargin-based AL can sometimes perform even worse than PL. While existing works\noffer different explanations in the low-dimensional regime, this paper shows\nthat the underlying mechanism is entirely different in high dimensions: we\nprove for logistic regression that PL outperforms margin-based AL even for\nnoiseless data and when using the Bayes optimal decision boundary for sampling.\nInsights from our proof indicate that this high-dimensional phenomenon is\nexacerbated when the separation between the classes is small. We corroborate\nthis intuition with experiments on 20 high-dimensional datasets spanning a\ndiverse range of applications, from finance and histology to chemistry and\ncomputer vision.\n", "date": "2022-12-01", "categories": []}
{"id": "2306.01277", "title": "Beyond Active Learning: Leveraging the Full Potential of Human\n  Interaction via Auto-Labeling, Human Correction, and Human Verification", "abstract": "  Active Learning (AL) is a human-in-the-loop framework to interactively and\nadaptively label data instances, thereby enabling significant gains in model\nperformance compared to random sampling. AL approaches function by selecting\nthe hardest instances to label, often relying on notions of diversity and\nuncertainty. However, we believe that these current paradigms of AL do not\nleverage the full potential of human interaction granted by automated label\nsuggestions. Indeed, we show that for many classification tasks and datasets,\nmost people verifying if an automatically suggested label is correct take\n$3\\times$ to $4\\times$ less time than they do changing an incorrect suggestion\nto the correct label (or labeling from scratch without any suggestion).\nUtilizing this result, we propose CLARIFIER (aCtive LeARnIng From tIEred\nhaRdness), an Interactive Learning framework that admits more effective use of\nhuman interaction by leveraging the reduced cost of verification. By targeting\nthe hard (uncertain) instances with existing AL methods, the intermediate\ninstances with a novel label suggestion scheme using submodular mutual\ninformation functions on a per-class basis, and the easy (confident) instances\nwith highest-confidence auto-labeling, CLARIFIER can improve over the\nperformance of existing AL approaches on multiple datasets -- particularly on\nthose that have a large number of classes -- by almost 1.5$\\times$ to 2$\\times$\nin terms of relative labeling cost.\n", "date": "2023-06-02", "categories": []}
{"id": "2305.19588", "title": "Active causal structure learning with advice", "abstract": "  We introduce the problem of active causal structure learning with advice. In\nthe typical well-studied setting, the learning algorithm is given the essential\ngraph for the observational distribution and is asked to recover the underlying\ncausal directed acyclic graph (DAG) $G^*$ while minimizing the number of\ninterventions made. In our setting, we are additionally given side information\nabout $G^*$ as advice, e.g. a DAG $G$ purported to be $G^*$. We ask whether the\nlearning algorithm can benefit from the advice when it is close to being\ncorrect, while still having worst-case guarantees even when the advice is\narbitrarily bad. Our work is in the same space as the growing body of research\non algorithms with predictions. When the advice is a DAG $G$, we design an\nadaptive search algorithm to recover $G^*$ whose intervention cost is at most\n$O(\\max\\{1, \\log \\psi\\})$ times the cost for verifying $G^*$; here, $\\psi$ is a\ndistance measure between $G$ and $G^*$ that is upper bounded by the number of\nvariables $n$, and is exactly 0 when $G=G^*$. Our approximation factor matches\nthe state-of-the-art for the advice-less setting.\n", "date": "2023-05-31", "categories": []}
{"id": "2305.19267", "title": "Parallelized Acquisition for Active Learning using Monte Carlo Sampling", "abstract": "  Bayesian inference remains one of the most important tool-kits for any\nscientist, but increasingly expensive likelihood functions are required for\never-more complex experiments, raising the cost of generating a Monte Carlo\nsample of the posterior. Recent attention has been directed towards the use of\nemulators of the posterior based on Gaussian Process (GP) regression combined\nwith active sampling to achieve comparable precision with far fewer costly\nlikelihood evaluations. Key to this approach is the batched acquisition of\nproposals, so that the true posterior can be evaluated in parallel. This is\nusually achieved via sequential maximization of the highly multimodal\nacquisition function. Unfortunately, this approach parallelizes poorly and is\nprone to getting stuck in local maxima. Our approach addresses this issue by\ngenerating nearly-optimal batches of candidates using an almost-embarrassingly\nparallel Nested Sampler on the mean prediction of the GP. The resulting\nnearly-sorted Monte Carlo sample is used to generate a batch of candidates\nranked according to their sequentially conditioned acquisition function values\nat little cost. The final sample can also be used for inferring marginal\nquantities. Our proposed implementation (NORA) demonstrates comparable accuracy\nto sequential conditioned acquisition optimization and efficient\nparallelization in various synthetic and cosmological inference problems.\n", "date": "2023-05-30", "categories": []}
{"id": "2305.17520", "title": "USIM-DAL: Uncertainty-aware Statistical Image Modeling-based Dense\n  Active Learning for Super-resolution", "abstract": "  Dense regression is a widely used approach in computer vision for tasks such\nas image super-resolution, enhancement, depth estimation, etc. However, the\nhigh cost of annotation and labeling makes it challenging to achieve accurate\nresults. We propose incorporating active learning into dense regression models\nto address this problem. Active learning allows models to select the most\ninformative samples for labeling, reducing the overall annotation cost while\nimproving performance. Despite its potential, active learning has not been\nwidely explored in high-dimensional computer vision regression tasks like\nsuper-resolution. We address this research gap and propose a new framework\ncalled USIM-DAL that leverages the statistical properties of colour images to\nlearn informative priors using probabilistic deep neural networks that model\nthe heteroscedastic predictive distribution allowing uncertainty\nquantification. Moreover, the aleatoric uncertainty from the network serves as\na proxy for error that is used for active learning. Our experiments on a wide\nvariety of datasets spanning applications in natural images (visual genome,\nBSD100), medical imaging (histopathology slides), and remote sensing (satellite\nimages) demonstrate the efficacy of the newly proposed USIM-DAL and superiority\nover several dense regression active learning methods.\n", "date": "2023-05-27", "categories": []}
{"id": "2305.17013", "title": "D-CALM: A Dynamic Clustering-based Active Learning Approach for\n  Mitigating Bias", "abstract": "  Despite recent advancements, NLP models continue to be vulnerable to bias.\nThis bias often originates from the uneven distribution of real-world data and\ncan propagate through the annotation process. Escalated integration of these\nmodels in our lives calls for methods to mitigate bias without overbearing\nannotation costs. While active learning (AL) has shown promise in training\nmodels with a small amount of annotated data, AL's reliance on the model's\nbehavior for selective sampling can lead to an accumulation of unwanted bias\nrather than bias mitigation. However, infusing clustering with AL can overcome\nthe bias issue of both AL and traditional annotation methods while exploiting\nAL's annotation efficiency. In this paper, we propose a novel adaptive\nclustering-based active learning algorithm, D-CALM, that dynamically adjusts\nclustering and annotation efforts in response to an estimated classifier\nerror-rate. Experiments on eight datasets for a diverse set of text\nclassification tasks, including emotion, hatespeech, dialog act, and book type\ndetection, demonstrate that our proposed algorithm significantly outperforms\nbaseline AL approaches with both pretrained transformers and traditional\nSupport Vector Machines. D-CALM showcases robustness against different measures\nof information gain and, as evident from our analysis of label and error\ndistribution, can significantly reduce unwanted model bias.\n", "date": "2023-05-26", "categories": []}
{"id": "2305.13342", "title": "On the Limitations of Simulating Active Learning", "abstract": "  Active learning (AL) is a human-and-model-in-the-loop paradigm that\niteratively selects informative unlabeled data for human annotation, aiming to\nimprove over random sampling. However, performing AL experiments with human\nannotations on-the-fly is a laborious and expensive process, thus unrealistic\nfor academic research. An easy fix to this impediment is to simulate AL, by\ntreating an already labeled and publicly available dataset as the pool of\nunlabeled data. In this position paper, we first survey recent literature and\nhighlight the challenges across all different steps within the AL loop. We\nfurther unveil neglected caveats in the experimental setup that can\nsignificantly affect the quality of AL research. We continue with an\nexploration of how the simulation setting can govern empirical findings,\narguing that it might be one of the answers behind the ever posed question\n``why do active learning algorithms sometimes fail to outperform random\nsampling?''. We argue that evaluating AL algorithms on available labeled\ndatasets might provide a lower bound as to their effectiveness in real data. We\nbelieve it is essential to collectively shape the best practices for AL\nresearch, particularly as engineering advancements in LLMs push the research\nfocus towards data-driven approaches (e.g., data efficiency, alignment,\nfairness). In light of this, we have developed guidelines for future work. Our\naim is to draw attention to these limitations within the community, in the hope\nof finding ways to address them.\n", "date": "2023-05-21", "categories": []}
{"id": "2305.10643", "title": "STREAMLINE: Streaming Active Learning for Realistic Multi-Distributional\n  Settings", "abstract": "  Deep neural networks have consistently shown great performance in several\nreal-world use cases like autonomous vehicles, satellite imaging, etc.,\neffectively leveraging large corpora of labeled training data. However,\nlearning unbiased models depends on building a dataset that is representative\nof a diverse range of realistic scenarios for a given task. This is challenging\nin many settings where data comes from high-volume streams, with each scenario\noccurring in random interleaved episodes at varying frequencies. We study\nrealistic streaming settings where data instances arrive in and are sampled\nfrom an episodic multi-distributional data stream. Using submodular information\nmeasures, we propose STREAMLINE, a novel streaming active learning framework\nthat mitigates scenario-driven slice imbalance in the working labeled data via\na three-step procedure of slice identification, slice-aware budgeting, and data\nselection. We extensively evaluate STREAMLINE on real-world streaming scenarios\nfor image classification and object detection tasks. We observe that STREAMLINE\nimproves the performance on infrequent yet critical slices of the data over\ncurrent baselines by up to $5\\%$ in terms of accuracy on our image\nclassification tasks and by up to $8\\%$ in terms of mAP on our object detection\ntasks.\n", "date": "2023-05-18", "categories": []}
{"id": "2305.07040", "title": "Sequential Experimental Design for Spectral Measurement: Active Learning\n  Using a Parametric Model", "abstract": "  In this study, we demonstrate a sequential experimental design for spectral\nmeasurements by active learning using parametric models as predictors. In\nspectral measurements, it is necessary to reduce the measurement time because\nof sample fragility and high energy costs. To improve the efficiency of\nexperiments, sequential experimental designs are proposed, in which the\nsubsequent measurement is designed by active learning using the data obtained\nbefore the measurement. Conventionally, parametric models are employed in data\nanalysis; when employed for active learning, they are expected to afford a\nsequential experimental design that improves the accuracy of data analysis.\nHowever, due to the complexity of the formulas, a sequential experimental\ndesign using general parametric models has not been realized. Therefore, we\napplied Bayesian inference-based data analysis using the exchange Monte Carlo\nmethod to realize a sequential experimental design with general parametric\nmodels. In this study, we evaluated the effectiveness of the proposed method by\napplying it to Bayesian spectral deconvolution and Bayesian Hamiltonian\nselection in X-ray photoelectron spectroscopy. Using numerical experiments with\nartificial data, we demonstrated that the proposed method improves the accuracy\nof model selection and parameter estimation while reducing the measurement time\ncompared with the results achieved without active learning or with active\nlearning using the Gaussian process regression.\n", "date": "2023-05-11", "categories": []}
{"id": "2211.01568", "title": "Fine-Tuning Language Models via Epistemic Neural Networks", "abstract": "  Language models often pre-train on large unsupervised text corpora, then\nfine-tune on additional task-specific data. However, typical fine-tuning\nschemes do not prioritize the examples that they tune on. We show that, if you\ncan prioritize informative training data, you can achieve better performance\nwhile using fewer labels. To do this we augment a language model with an\nepinet: a small additional network that helps to estimate model uncertainty and\nforms an \\textit{epistemic neural network} (ENN). ENNs are neural networks that\ncan know what they don't know. Using an epinet to prioritize uncertain data, we\ncan fine-tune BERT on GLUE tasks to the same performance while using 2x less\ndata than training without prioritization. We also investigate performance in\nsynthetic neural network generative models designed to build understanding. In\neach setting, using an epinet outperforms heuristic active learning schemes.\n", "date": "2022-11-03", "categories": []}
{"id": "2209.06995", "title": "Cold-Start Data Selection for Few-shot Language Model Fine-tuning: A\n  Prompt-Based Uncertainty Propagation Approach", "abstract": "  Large Language Models have demonstrated remarkable few-shot performance, but\nthe performance can be sensitive to the selection of few-shot instances. We\npropose PATRON, a new method that uses prompt-based uncertainty estimation for\ndata selection for pre-trained language model fine-tuning under cold-start\nscenarios, i.e., no initial labeled data are available. In PATRON, we design\n(1) a prompt-based uncertainty propagation approach to estimate the importance\nof data points and (2) a partition-then-rewrite (PTR) strategy to promote\nsample diversity when querying for annotations. Experiments on six text\nclassification datasets show that PATRON outperforms the strongest cold-start\ndata selection baselines by up to 6.9%. Besides, with 128 labels only, PATRON\nachieves 91.0% and 92.1% of the fully supervised performance based on vanilla\nfine-tuning and prompt-based learning respectively. Our implementation of\nPATRON is available at \\url{https://github.com/yueyu1030/Patron}.\n", "date": "2022-09-15", "categories": []}
{"id": "2305.02459", "title": "Transfer and Active Learning for Dissonance Detection: Addressing the\n  Rare-Class Challenge", "abstract": "  While transformer-based systems have enabled greater accuracies with fewer\ntraining examples, data acquisition obstacles still persist for rare-class\ntasks -- when the class label is very infrequent (e.g. < 5% of samples). Active\nlearning has in general been proposed to alleviate such challenges, but choice\nof selection strategy, the criteria by which rare-class examples are chosen,\nhas not been systematically evaluated. Further, transformers enable iterative\ntransfer-learning approaches. We propose and investigate transfer- and active\nlearning solutions to the rare class problem of dissonance detection through\nutilizing models trained on closely related tasks and the evaluation of\nacquisition strategies, including a proposed probability-of-rare-class (PRC)\napproach. We perform these experiments for a specific rare class problem:\ncollecting language samples of cognitive dissonance from social media. We find\nthat PRC is a simple and effective strategy to guide annotations and ultimately\nimprove model accuracy while transfer-learning in a specific order can improve\nthe cold-start performance of the learner but does not benefit iterations of\nactive learning.\n", "date": "2023-05-03", "categories": []}
{"id": "2305.00111", "title": "Active Reinforcement Learning for Personalized Stress Monitoring in\n  Everyday Settings", "abstract": "  Most existing sensor-based monitoring frameworks presume that a large\navailable labeled dataset is processed to train accurate detection models.\nHowever, in settings where personalization is necessary at deployment time to\nfine-tune the model, a person-specific dataset needs to be collected online by\ninteracting with the users. Optimizing the collection of labels in such phase\nis instrumental to impose a tolerable burden on the users while maximizing\npersonal improvement. In this paper, we consider a fine-grain stress detection\nproblem based on wearable sensors targeting everyday settings, and propose a\nnovel context-aware active learning strategy capable of jointly maximizing the\nmeaningfulness of the signal samples we request the user to label and the\nresponse rate. We develop a multilayered sensor-edge-cloud platform to\nperiodically capture physiological signals and process them in real-time, as\nwell as to collect labels and retrain the detection model. We collect a large\ndataset and show that the context-aware active learning technique we propose\nachieves a desirable detection performance using 88\\% and 32\\% fewer queries\nfrom users compared to a randomized strategy and a traditional active learning\nstrategy, respectively.\n", "date": "2023-04-28", "categories": []}
{"id": "2207.12710", "title": "Active Learning of Ordinal Embeddings: A User Study on Football Data", "abstract": "  Humans innately measure distance between instances in an unlabeled dataset\nusing an unknown similarity function. Distance metrics can only serve as proxy\nfor similarity in information retrieval of similar instances. Learning a good\nsimilarity function from human annotations improves the quality of retrievals.\nThis work uses deep metric learning to learn these user-defined similarity\nfunctions from few annotations for a large football trajectory dataset. We\nadapt an entropy-based active learning method with recent work from triplet\nmining to collect easy-to-answer but still informative annotations from human\nparticipants and use them to train a deep convolutional network that\ngeneralizes to unseen samples. Our user study shows that our approach improves\nthe quality of the information retrieval compared to a previous deep metric\nlearning approach that relies on a Siamese network. Specifically, we shed light\non the strengths and weaknesses of passive sampling heuristics and active\nlearners alike by analyzing the participants' response efficacy. To this end,\nwe collect accuracy, algorithmic time complexity, the participants' fatigue and\ntime-to-response, qualitative self-assessment and statements, as well as the\neffects of mixed-expertise annotators and their consistency on model\nperformance and transfer-learning.\n", "date": "2022-07-26", "categories": []}
{"id": "2304.11989", "title": "Generative Flow Networks for Precise Reward-Oriented Active Learning on\n  Graphs", "abstract": "  Many score-based active learning methods have been successfully applied to\ngraph-structured data, aiming to reduce the number of labels and achieve better\nperformance of graph neural networks based on predefined score functions.\nHowever, these algorithms struggle to learn policy distributions that are\nproportional to rewards and have limited exploration capabilities. In this\npaper, we innovatively formulate the graph active learning problem as a\ngenerative process, named GFlowGNN, which generates various samples through\nsequential actions with probabilities precisely proportional to a predefined\nreward function. Furthermore, we propose the concept of flow nodes and flow\nfeatures to efficiently model graphs as flows based on generative flow\nnetworks, where the policy network is trained with specially designed rewards.\nExtensive experiments on real datasets show that the proposed approach has good\nexploration capability and transferability, outperforming various\nstate-of-the-art methods.\n", "date": "2023-04-24", "categories": []}
{"id": "2304.07445", "title": "A framework for fully autonomous design of materials via multiobjective\n  optimization and active learning: challenges and next steps", "abstract": "  In order to deploy machine learning in a real-world self-driving laboratory\nwhere data acquisition is costly and there are multiple competing design\ncriteria, systems need to be able to intelligently sample while balancing\nperformance trade-offs and constraints. For these reasons, we present an active\nlearning process based on multiobjective black-box optimization with\ncontinuously updated machine learning models. This workflow is built on\nopen-source technologies for real-time data streaming and modular\nmultiobjective optimization software development. We demonstrate a proof of\nconcept for this workflow through the autonomous operation of a continuous-flow\nchemistry laboratory, which identifies ideal manufacturing conditions for the\nelectrolyte 2,2,2-trifluoroethyl methyl carbonate.\n", "date": "2023-04-15", "categories": []}
{"id": "2304.08151", "title": "Prediction-Oriented Bayesian Active Learning", "abstract": "  Information-theoretic approaches to active learning have traditionally\nfocused on maximising the information gathered about the model parameters, most\ncommonly by optimising the BALD score. We highlight that this can be suboptimal\nfrom the perspective of predictive performance. For example, BALD lacks a\nnotion of an input distribution and so is prone to prioritise data of limited\nrelevance. To address this we propose the expected predictive information gain\n(EPIG), an acquisition function that measures information gain in the space of\npredictions rather than parameters. We find that using EPIG leads to stronger\npredictive performance compared with BALD across a range of datasets and\nmodels, and thus provides an appealing drop-in replacement.\n", "date": "2023-04-17", "categories": []}
{"id": "2304.07031", "title": "Spectral Transfer Guided Active Domain Adaptation For Thermal Imagery", "abstract": "  The exploitation of visible spectrum datasets has led deep networks to show\nremarkable success. However, real-world tasks include low-lighting conditions\nwhich arise performance bottlenecks for models trained on large-scale RGB image\ndatasets. Thermal IR cameras are more robust against such conditions.\nTherefore, the usage of thermal imagery in real-world applications can be\nuseful. Unsupervised domain adaptation (UDA) allows transferring information\nfrom a source domain to a fully unlabeled target domain. Despite substantial\nimprovements in UDA, the performance gap between UDA and its supervised\nlearning counterpart remains significant. By picking a small number of target\nsamples to annotate and using them in training, active domain adaptation tries\nto mitigate this gap with minimum annotation expense. We propose an active\ndomain adaptation method in order to examine the efficiency of combining the\nvisible spectrum and thermal imagery modalities. When the domain gap is\nconsiderably large as in the visible-to-thermal task, we may conclude that the\nmethods without explicit domain alignment cannot achieve their full potential.\nTo this end, we propose a spectral transfer guided active domain adaptation\nmethod to select the most informative unlabeled target samples while aligning\nsource and target domains. We used the large-scale visible spectrum dataset\nMS-COCO as the source domain and the thermal dataset FLIR ADAS as the target\ndomain to present the results of our method. Extensive experimental evaluation\ndemonstrates that our proposed method outperforms the state-of-the-art active\ndomain adaptation methods. The code and models are publicly available.\n", "date": "2023-04-14", "categories": []}
{"id": "2304.06277", "title": "Optimizing Multi-Domain Performance with Active Learning-based\n  Improvement Strategies", "abstract": "  Improving performance in multiple domains is a challenging task, and often\nrequires significant amounts of data to train and test models. Active learning\ntechniques provide a promising solution by enabling models to select the most\ninformative samples for labeling, thus reducing the amount of labeled data\nrequired to achieve high performance. In this paper, we present an active\nlearning-based framework for improving performance across multiple domains. Our\napproach consists of two stages: first, we use an initial set of labeled data\nto train a base model, and then we iteratively select the most informative\nsamples for labeling to refine the model. We evaluate our approach on several\nmulti-domain datasets, including image classification, sentiment analysis, and\nobject recognition. Our experiments demonstrate that our approach consistently\noutperforms baseline methods and achieves state-of-the-art performance on\nseveral datasets. We also show that our method is highly efficient, requiring\nsignificantly fewer labeled samples than other active learning-based methods.\nOverall, our approach provides a practical and effective solution for improving\nperformance across multiple domains using active learning techniques.\n", "date": "2023-04-13", "categories": []}
{"id": "2304.05578", "title": "Does Informativeness Matter? Active Learning for Educational Dialogue\n  Act Classification", "abstract": "  Dialogue Acts (DAs) can be used to explain what expert tutors do and what\nstudents know during the tutoring process. Most empirical studies adopt the\nrandom sampling method to obtain sentence samples for manual annotation of DAs,\nwhich are then used to train DA classifiers. However, these studies have paid\nlittle attention to sample informativeness, which can reflect the information\nquantity of the selected samples and inform the extent to which a classifier\ncan learn patterns. Notably, the informativeness level may vary among the\nsamples and the classifier might only need a small amount of low informative\nsamples to learn the patterns. Random sampling may overlook sample\ninformativeness, which consumes human labelling costs and contributes less to\ntraining the classifiers. As an alternative, researchers suggest employing\nstatistical sampling methods of Active Learning (AL) to identify the\ninformative samples for training the classifiers. However, the use of AL\nmethods in educational DA classification tasks is under-explored. In this\npaper, we examine the informativeness of annotated sentence samples. Then, the\nstudy investigates how the AL methods can select informative samples to support\nDA classifiers in the AL sampling process. The results reveal that most\nannotated sentences present low informativeness in the training dataset and the\npatterns of these sentences can be easily captured by the DA classifier. We\nalso demonstrate how AL methods can reduce the cost of manual annotation in the\nAL sampling process.\n", "date": "2023-04-12", "categories": []}
{"id": "2211.07881", "title": "ET-AL: Entropy-Targeted Active Learning for Bias Mitigation in Materials\n  Data", "abstract": "  Growing materials data and data-driven informatics drastically promote the\ndiscovery and design of materials. While there are significant advancements in\ndata-driven models, the quality of data resources is less studied despite its\nhuge impact on model performance. In this work, we focus on data bias arising\nfrom uneven coverage of materials families in existing knowledge. Observing\ndifferent diversities among crystal systems in common materials databases, we\npropose an information entropy-based metric for measuring this bias. To\nmitigate the bias, we develop an entropy-targeted active learning (ET-AL)\nframework, which guides the acquisition of new data to improve the diversity of\nunderrepresented crystal systems. We demonstrate the capability of ET-AL for\nbias mitigation and the resulting improvement in downstream machine learning\nmodels. This approach is broadly applicable to data-driven materials discovery,\nincluding autonomous data acquisition and dataset trimming to reduce bias, as\nwell as data-driven informatics in other scientific domains.\n", "date": "2022-11-15", "categories": []}
{"id": "2304.05246", "title": "OpenAL: Evaluation and Interpretation of Active Learning Strategies", "abstract": "  Despite the vast body of literature on Active Learning (AL), there is no\ncomprehensive and open benchmark allowing for efficient and simple comparison\nof proposed samplers. Additionally, the variability in experimental settings\nacross the literature makes it difficult to choose a sampling strategy, which\nis critical due to the one-off nature of AL experiments. To address those\nlimitations, we introduce OpenAL, a flexible and open-source framework to\neasily run and compare sampling AL strategies on a collection of realistic\ntasks. The proposed benchmark is augmented with interpretability metrics and\nstatistical analysis methods to understand when and why some samplers\noutperform others. Last but not least, practitioners can easily extend the\nbenchmark by submitting their own AL samplers.\n", "date": "2023-04-11", "categories": []}
{"id": "2212.13067", "title": "Online Active Learning for Soft Sensor Development using Semi-Supervised\n  Autoencoders", "abstract": "  Data-driven soft sensors are extensively used in industrial and chemical\nprocesses to predict hard-to-measure process variables whose real value is\ndifficult to track during routine operations. The regression models used by\nthese sensors often require a large number of labeled examples, yet obtaining\nthe label information can be very expensive given the high time and cost\nrequired by quality inspections. In this context, active learning methods can\nbe highly beneficial as they can suggest the most informative labels to query.\nHowever, most of the active learning strategies proposed for regression focus\non the offline setting. In this work, we adapt some of these approaches to the\nstream-based scenario and show how they can be used to select the most\ninformative data points. We also demonstrate how to use a semi-supervised\narchitecture based on orthogonal autoencoders to learn salient features in a\nlower dimensional space. The Tennessee Eastman Process is used to compare the\npredictive performance of the proposed approaches.\n", "date": "2022-12-26", "categories": []}
{"id": "2304.02484", "title": "A dynamic Bayesian optimized active recommender system for\n  curiosity-driven Human-in-the-loop automated experiments", "abstract": "  Optimization of experimental materials synthesis and characterization through\nactive learning methods has been growing over the last decade, with examples\nranging from measurements of diffraction on combinatorial alloys at\nsynchrotrons, to searches through chemical space with automated synthesis\nrobots for perovskites. In virtually all cases, the target property of interest\nfor optimization is defined apriori with limited human feedback during\noperation. In contrast, here we present the development of a new type of human\nin the loop experimental workflow, via a Bayesian optimized active recommender\nsystem (BOARS), to shape targets on the fly, employing human feedback. We\nshowcase examples of this framework applied to pre-acquired piezoresponse force\nspectroscopy of a ferroelectric thin film, and then implement this in real time\non an atomic force microscope, where the optimization proceeds to find\nsymmetric piezoresponse amplitude hysteresis loops. It is found that such\nfeatures appear more affected by subsurface defects than the local domain\nstructure. This work shows the utility of human-augmented machine learning\napproaches for curiosity-driven exploration of systems across experimental\ndomains. The analysis reported here is summarized in Colab Notebook for the\npurpose of tutorial and application to other data:\nhttps://github.com/arpanbiswas52/varTBO\n", "date": "2023-04-05", "categories": []}
{"id": "2304.01404", "title": "Adaptive Defective Area Identification in Material Surface Using Active\n  Transfer Learning-based Level Set Estimation", "abstract": "  In material characterization, identifying defective areas on a material\nsurface is fundamental. The conventional approach involves measuring the\nrelevant physical properties point-by-point at the predetermined mesh grid\npoints on the surface and determining the area at which the property does not\nreach the desired level. To identify defective areas more efficiently, we\npropose adaptive mapping methods in which measurement resources are used\npreferentially to detect the boundaries of defective areas. We interpret this\nproblem as an active-learning (AL) of the level set estimation (LSE) problem.\nThe goal of AL-based LSE is to determine the level set of the physical property\nfunction defined on the surface with as small number of measurements as\npossible. Furthermore, to handle the situations in which materials with similar\nspecifications are repeatedly produced, we introduce a transfer learning\napproach so that the information of previously produced materials can be\neffectively utilized. As a proof-of-concept, we applied the proposed methods to\nthe red-zone estimation problem of silicon wafers and demonstrated that we\ncould identify the defective areas with significantly lower measurement costs\nthan those of conventional methods.\n", "date": "2023-04-03", "categories": []}
{"id": "2303.16637", "title": "MuRAL: Multi-Scale Region-based Active Learning for Object Detection", "abstract": "  Obtaining large-scale labeled object detection dataset can be costly and\ntime-consuming, as it involves annotating images with bounding boxes and class\nlabels. Thus, some specialized active learning methods have been proposed to\nreduce the cost by selecting either coarse-grained samples or fine-grained\ninstances from unlabeled data for labeling. However, the former approaches\nsuffer from redundant labeling, while the latter methods generally lead to\ntraining instability and sampling bias. To address these challenges, we propose\na novel approach called Multi-scale Region-based Active Learning (MuRAL) for\nobject detection. MuRAL identifies informative regions of various scales to\nreduce annotation costs for well-learned objects and improve training\nperformance. The informative region score is designed to consider both the\npredicted confidence of instances and the distribution of each object category,\nenabling our method to focus more on difficult-to-detect classes. Moreover,\nMuRAL employs a scale-aware selection strategy that ensures diverse regions are\nselected from different scales for labeling and downstream finetuning, which\nenhances training stability. Our proposed method surpasses all existing\ncoarse-grained and fine-grained baselines on Cityscapes and MS COCO datasets,\nand demonstrates significant improvement in difficult category performance.\n", "date": "2023-03-29", "categories": []}
{"id": "2303.14382", "title": "Active Finetuning: Exploiting Annotation Budget in the\n  Pretraining-Finetuning Paradigm", "abstract": "  Given the large-scale data and the high annotation cost,\npretraining-finetuning becomes a popular paradigm in multiple computer vision\ntasks. Previous research has covered both the unsupervised pretraining and\nsupervised finetuning in this paradigm, while little attention is paid to\nexploiting the annotation budget for finetuning. To fill in this gap, we\nformally define this new active finetuning task focusing on the selection of\nsamples for annotation in the pretraining-finetuning paradigm. We propose a\nnovel method called ActiveFT for active finetuning task to select a subset of\ndata distributing similarly with the entire unlabeled pool and maintaining\nenough diversity by optimizing a parametric model in the continuous space. We\nprove that the Earth Mover's distance between the distributions of the selected\nsubset and the entire data pool is also reduced in this process. Extensive\nexperiments show the leading performance and high efficiency of ActiveFT\nsuperior to baselines on both image classification and semantic segmentation.\nOur code is released at https://github.com/yichen928/ActiveFT.\n", "date": "2023-03-25", "categories": []}
{"id": "2303.14433", "title": "Deep Active Learning with Contrastive Learning Under Realistic Data Pool\n  Assumptions", "abstract": "  Active learning aims to identify the most informative data from an unlabeled\ndata pool that enables a model to reach the desired accuracy rapidly. This\nbenefits especially deep neural networks which generally require a huge number\nof labeled samples to achieve high performance. Most existing active learning\nmethods have been evaluated in an ideal setting where only samples relevant to\nthe target task, i.e., in-distribution samples, exist in an unlabeled data\npool. A data pool gathered from the wild, however, is likely to include samples\nthat are irrelevant to the target task at all and/or too ambiguous to assign a\nsingle class label even for the oracle. We argue that assuming an unlabeled\ndata pool consisting of samples from various distributions is more realistic.\nIn this work, we introduce new active learning benchmarks that include\nambiguous, task-irrelevant out-of-distribution as well as in-distribution\nsamples. We also propose an active learning method designed to acquire\ninformative in-distribution samples in priority. The proposed method leverages\nboth labeled and unlabeled data pools and selects samples from clusters on the\nfeature space constructed via contrastive learning. Experimental results\ndemonstrate that the proposed method requires a lower annotation budget than\nexisting active learning methods to reach the same level of accuracy.\n", "date": "2023-03-25", "categories": []}
{"id": "2303.00141", "title": "Containing a spread through sequential learning: to exploit or to\n  explore?", "abstract": "  The spread of an undesirable contact process, such as an infectious disease\n(e.g. COVID-19), is contained through testing and isolation of infected nodes.\nThe temporal and spatial evolution of the process (along with containment\nthrough isolation) render such detection as fundamentally different from active\nsearch detection strategies. In this work, through an active learning approach,\nwe design testing and isolation strategies to contain the spread and minimize\nthe cumulative infections under a given test budget. We prove that the\nobjective can be optimized, with performance guarantees, by greedily selecting\nthe nodes to test. We further design reward-based methodologies that\neffectively minimize an upper bound on the cumulative infections and are\ncomputationally more tractable in large networks. These policies, however, need\nknowledge about the nodes' infection probabilities which are dynamically\nchanging and have to be learned by sequential testing. We develop a\nmessage-passing framework for this purpose and, building on that, show novel\ntradeoffs between exploitation of knowledge through reward-based heuristics and\nexploration of the unknown through a carefully designed probabilistic testing.\nThe tradeoffs are fundamentally distinct from the classical counterparts under\nactive search or multi-armed bandit problems (MABs). We provably show the\nnecessity of exploration in a stylized network and show through simulations\nthat exploration can outperform exploitation in various synthetic and real-data\nnetworks depending on the parameters of the network and the spread.\n", "date": "2023-03-01", "categories": []}
{"id": "2303.13089", "title": "Box-Level Active Detection", "abstract": "  Active learning selects informative samples for annotation within budget,\nwhich has proven efficient recently on object detection. However, the widely\nused active detection benchmarks conduct image-level evaluation, which is\nunrealistic in human workload estimation and biased towards crowded images.\nFurthermore, existing methods still perform image-level annotation, but equally\nscoring all targets within the same image incurs waste of budget and redundant\nlabels. Having revealed above problems and limitations, we introduce a\nbox-level active detection framework that controls a box-based budget per\ncycle, prioritizes informative targets and avoids redundancy for fair\ncomparison and efficient application.\n  Under the proposed box-level setting, we devise a novel pipeline, namely\nComplementary Pseudo Active Strategy (ComPAS). It exploits both human\nannotations and the model intelligence in a complementary fashion: an efficient\ninput-end committee queries labels for informative objects only; meantime\nwell-learned targets are identified by the model and compensated with\npseudo-labels. ComPAS consistently outperforms 10 competitors under 4 settings\nin a unified codebase. With supervision from labeled data only, it achieves\n100% supervised performance of VOC0712 with merely 19% box annotations. On the\nCOCO dataset, it yields up to 4.3% mAP improvement over the second-best method.\nComPAS also supports training with the unlabeled pool, where it surpasses 90%\nCOCO supervised performance with 85% label reduction. Our source code is\npublicly available at https://github.com/lyumengyao/blad.\n", "date": "2023-03-23", "categories": []}
{"id": "2303.12317", "title": "Re-thinking Federated Active Learning based on Inter-class Diversity", "abstract": "  Although federated learning has made awe-inspiring advances, most studies\nhave assumed that the client's data are fully labeled. However, in a real-world\nscenario, every client may have a significant amount of unlabeled instances.\nAmong the various approaches to utilizing unlabeled data, a federated active\nlearning framework has emerged as a promising solution. In the decentralized\nsetting, there are two types of available query selector models, namely\n'global' and 'local-only' models, but little literature discusses their\nperformance dominance and its causes. In this work, we first demonstrate that\nthe superiority of two selector models depends on the global and local\ninter-class diversity. Furthermore, we observe that the global and local-only\nmodels are the keys to resolving the imbalance of each side. Based on our\nfindings, we propose LoGo, a FAL sampling strategy robust to varying local\nheterogeneity levels and global imbalance ratio, that integrates both models by\ntwo steps of active selection scheme. LoGo consistently outperforms six active\nlearning strategies in the total number of 38 experimental settings.\n", "date": "2023-03-22", "categories": []}
{"id": "2303.12760", "title": "Uncertainty Aware Active Learning for Reconfiguration of Pre-trained\n  Deep Object-Detection Networks for New Target Domains", "abstract": "  Object detection is one of the most important and fundamental aspects of\ncomputer vision tasks, which has been broadly utilized in pose estimation,\nobject tracking and instance segmentation models. To obtain training data for\nobject detection model efficiently, many datasets opt to obtain their\nunannotated data in video format and the annotator needs to draw a bounding box\naround each object in the images. Annotating every frame from a video is costly\nand inefficient since many frames contain very similar information for the\nmodel to learn from. How to select the most informative frames from a video to\nannotate has become a highly practical task to solve but attracted little\nattention in research. In this paper, we proposed a novel active learning\nalgorithm for object detection models to tackle this problem. In the proposed\nactive learning algorithm, both classification and localization informativeness\nof unlabelled data are measured and aggregated. Utilizing the temporal\ninformation from video frames, two novel localization informativeness\nmeasurements are proposed. Furthermore, a weight curve is proposed to avoid\nquerying adjacent frames. Proposed active learning algorithm with multiple\nconfigurations was evaluated on the MuPoTS dataset and FootballPD dataset.\n", "date": "2023-03-22", "categories": []}
{"id": "2303.11678", "title": "Full or Weak annotations? An adaptive strategy for budget-constrained\n  annotation campaigns", "abstract": "  Annotating new datasets for machine learning tasks is tedious,\ntime-consuming, and costly. For segmentation applications, the burden is\nparticularly high as manual delineations of relevant image content are often\nextremely expensive or can only be done by experts with domain-specific\nknowledge. Thanks to developments in transfer learning and training with weak\nsupervision, segmentation models can now also greatly benefit from annotations\nof different kinds. However, for any new domain application looking to use weak\nsupervision, the dataset builder still needs to define a strategy to distribute\nfull segmentation and other weak annotations. Doing so is challenging, however,\nas it is a priori unknown how to distribute an annotation budget for a given\nnew dataset. To this end, we propose a novel approach to determine annotation\nstrategies for segmentation datasets, whereby estimating what proportion of\nsegmentation and classification annotations should be collected given a fixed\nbudget. To do so, our method sequentially determines proportions of\nsegmentation and classification annotations to collect for budget-fractions by\nmodeling the expected improvement of the final segmentation model. We show in\nour experiments that our approach yields annotations that perform very close to\nthe optimal for a number of different annotation budgets and datasets.\n", "date": "2023-03-21", "categories": []}
{"id": "2302.04075", "title": "Best Practices in Active Learning for Semantic Segmentation", "abstract": "  Active learning is particularly of interest for semantic segmentation, where\nannotations are costly. Previous academic studies focused on datasets that are\nalready very diverse and where the model is trained in a supervised manner with\na large annotation budget. In contrast, data collected in many driving\nscenarios is highly redundant, and most medical applications are subject to\nvery constrained annotation budgets. This work investigates the various types\nof existing active learning methods for semantic segmentation under diverse\nconditions across three dimensions - data distribution w.r.t. different\nredundancy levels, integration of semi-supervised learning, and different\nlabeling budgets. We find that these three underlying factors are decisive for\nthe selection of the best active learning approach. As an outcome of our study,\nwe provide a comprehensive usage guide to obtain the best performance for each\ncase. We also propose an exemplary evaluation task for driving scenarios, where\ndata has high redundancy, to showcase the practical implications of our\nresearch findings.\n", "date": "2023-02-08", "categories": []}
{"id": "2303.08978", "title": "Active Semi-Supervised Learning by Exploring Per-Sample Uncertainty and\n  Consistency", "abstract": "  Active Learning (AL) and Semi-supervised Learning are two techniques that\nhave been studied to reduce the high cost of deep learning by using a small\namount of labeled data and a large amount of unlabeled data. To improve the\naccuracy of models at a lower cost, we propose a method called Active\nSemi-supervised Learning (ASSL), which combines AL and SSL. To maximize the\nsynergy between AL and SSL, we focused on the differences between ASSL and AL.\nASSL involves more dynamic model updates than AL due to the use of unlabeled\ndata in the training process, resulting in the temporal instability of the\npredicted probabilities of the unlabeled data. This makes it difficult to\ndetermine the true uncertainty of the unlabeled data in ASSL. To address this,\nwe adopted techniques such as exponential moving average (EMA) and upper\nconfidence bound (UCB) used in reinforcement learning. Additionally, we\nanalyzed the effect of label noise on unsupervised learning by using weak and\nstrong augmentation pairs to address datainconsistency. By considering both\nuncertainty and datainconsistency, we acquired data samples that were used in\nthe proposed ASSL method. Our experiments showed that ASSL achieved about 5.3\ntimes higher computational efficiency than SSL while achieving the same\nperformance, and it outperformed the state-of-the-art AL method.\n", "date": "2023-03-15", "categories": []}
{"id": "2207.10018", "title": "Mitigating Algorithmic Bias with Limited Annotations", "abstract": "  Existing work on fairness modeling commonly assumes that sensitive attributes\nfor all instances are fully available, which may not be true in many real-world\napplications due to the high cost of acquiring sensitive information. When\nsensitive attributes are not disclosed or available, it is needed to manually\nannotate a small part of the training data to mitigate bias. However, the\nskewed distribution across different sensitive groups preserves the skewness of\nthe original dataset in the annotated subset, which leads to non-optimal bias\nmitigation. To tackle this challenge, we propose Active Penalization Of\nDiscrimination (APOD), an interactive framework to guide the limited\nannotations towards maximally eliminating the effect of algorithmic bias. The\nproposed APOD integrates discrimination penalization with active instance\nselection to efficiently utilize the limited annotation budget, and it is\ntheoretically proved to be capable of bounding the algorithmic bias. According\nto the evaluation on five benchmark datasets, APOD outperforms the\nstate-of-the-arts baseline methods under the limited annotation budget, and\nshows comparable performance to fully annotated bias mitigation, which\ndemonstrates that APOD could benefit real-world applications when sensitive\ninformation is limited.\n", "date": "2022-07-20", "categories": []}
{"id": "2303.05225", "title": "Active Learning Based Domain Adaptation for Tissue Segmentation of\n  Histopathological Images", "abstract": "  Accurate segmentation of tissue in histopathological images can be very\nbeneficial for defining regions of interest (ROI) for streamline of diagnostic\nand prognostic tasks. Still, adapting to different domains is essential for\nhistopathology image analysis, as the visual characteristics of tissues can\nvary significantly across datasets. Yet, acquiring sufficient annotated data in\nthe medical domain is cumbersome and time-consuming. The labeling effort can be\nsignificantly reduced by leveraging active learning, which enables the\nselective annotation of the most informative samples. Our proposed method\nallows for fine-tuning a pre-trained deep neural network using a small set of\nlabeled data from the target domain, while also actively selecting the most\ninformative samples to label next. We demonstrate that our approach performs\nwith significantly fewer labeled samples compared to traditional supervised\nlearning approaches for similar F1-scores, using barely a 59\\% of the training\nset. We also investigate the distribution of class balance to establish\nannotation guidelines.\n", "date": "2023-03-09", "categories": []}
{"id": "2303.02721", "title": "Active learning using region-based sampling", "abstract": "  We present a general-purpose active learning scheme for data in metric\nspaces. The algorithm maintains a collection of neighborhoods of different\nsizes and uses label queries to identify those that have a strong bias towards\none particular label; when two such neighborhoods intersect and have different\nlabels, the region of overlap is treated as a ``known unknown'' and is a target\nof future active queries. We give label complexity bounds for this method that\ndo not rely on assumptions about the data and we instantiate them in several\ncases of interest.\n", "date": "2023-03-05", "categories": []}
{"id": "2303.00870", "title": "Implementing Active Learning in Cybersecurity: Detecting Anomalies in\n  Redacted Emails", "abstract": "  Research on email anomaly detection has typically relied on specially\nprepared datasets that may not adequately reflect the type of data that occurs\nin industry settings. In our research, at a major financial services company,\nprivacy concerns prevented inspection of the bodies of emails and attachment\ndetails (although subject headings and attachment filenames were available).\nThis made labeling possible anomalies in the resulting redacted emails more\ndifficult. Another source of difficulty is the high volume of emails combined\nwith the scarcity of resources making machine learning (ML) a necessity, but\nalso creating a need for more efficient human training of ML models. Active\nlearning (AL) has been proposed as a way to make human training of ML models\nmore efficient. However, the implementation of Active Learning methods is a\nhuman-centered AI challenge due to potential human analyst uncertainty, and the\nlabeling task can be further complicated in domains such as the cybersecurity\ndomain (or healthcare, aviation, etc.) where mistakes in labeling can have\nhighly adverse consequences. In this paper we present research results\nconcerning the application of Active Learning to anomaly detection in redacted\nemails, comparing the utility of different methods for implementing active\nlearning in this context. We evaluate different AL strategies and their impact\non resulting model performance. We also examine how ratings of confidence that\nexperts have in their labels can inform AL. The results obtained are discussed\nin terms of their implications for AL methodology and for the role of experts\nin model-assisted email anomaly screening.\n", "date": "2023-03-01", "categories": []}
{"id": "2303.00894", "title": "Active Reward Learning from Multiple Teachers", "abstract": "  Reward learning algorithms utilize human feedback to infer a reward function,\nwhich is then used to train an AI system. This human feedback is often a\npreference comparison, in which the human teacher compares several samples of\nAI behavior and chooses which they believe best accomplishes the objective.\nWhile reward learning typically assumes that all feedback comes from a single\nteacher, in practice these systems often query multiple teachers to gather\nsufficient training data. In this paper, we investigate this disparity, and\nfind that algorithmic evaluation of these different sources of feedback\nfacilitates more accurate and efficient reward learning. We formally analyze\nthe value of information (VOI) when reward learning from teachers with varying\nlevels of rationality, and define and evaluate an algorithm that utilizes this\nVOI to actively select teachers to query for feedback. Surprisingly, we find\nthat it is often more informative to query comparatively irrational teachers.\nBy formalizing this problem and deriving an analytical solution, we hope to\nfacilitate improvement in reward learning approaches to aligning AI behavior\nwith human values.\n", "date": "2023-03-02", "categories": []}
{"id": "2302.14567", "title": "Active Learning with Combinatorial Coverage", "abstract": "  Active learning is a practical field of machine learning that automates the\nprocess of selecting which data to label. Current methods are effective in\nreducing the burden of data labeling but are heavily model-reliant. This has\nled to the inability of sampled data to be transferred to new models as well as\nissues with sampling bias. Both issues are of crucial concern in machine\nlearning deployment. We propose active learning methods utilizing combinatorial\ncoverage to overcome these issues. The proposed methods are data-centric, as\nopposed to model-centric, and through our experiments we show that the\ninclusion of coverage in active learning leads to sampling data that tends to\nbe the best in transferring to better performing models and has a competitive\nsampling bias compared to benchmark methods.\n", "date": "2023-02-28", "categories": []}
{"id": "2302.13960", "title": "Acquisition Conditioned Oracle for Nongreedy Active Feature Acquisition", "abstract": "  We develop novel methodology for active feature acquisition (AFA), the study\nof how to sequentially acquire a dynamic (on a per instance basis) subset of\nfeatures that minimizes acquisition costs whilst still yielding accurate\npredictions. The AFA framework can be useful in a myriad of domains, including\nhealth care applications where the cost of acquiring additional features for a\npatient (in terms of time, money, risk, etc.) can be weighed against the\nexpected improvement to diagnostic performance. Previous approaches for AFA\nhave employed either: deep learning RL techniques, which have difficulty\ntraining policies in the AFA MDP due to sparse rewards and a complicated action\nspace; deep learning surrogate generative models, which require modeling\ncomplicated multidimensional conditional distributions; or greedy policies,\nwhich fail to account for how joint feature acquisitions can be informative\ntogether for better predictions. In this work we show that we can bypass many\nof these challenges with a novel, nonparametric oracle based approach, which we\ncoin the acquisition conditioned oracle (ACO). Extensive experiments show the\nsuperiority of the ACO to state-of-the-art AFA methods when acquiring features\nfor both predictions and general decision-making.\n", "date": "2023-02-27", "categories": []}
{"id": "2302.12667", "title": "Deep active learning for nonlinear system identification", "abstract": "  The exploding research interest for neural networks in modeling nonlinear\ndynamical systems is largely explained by the networks' capacity to model\ncomplex input-output relations directly from data. However, they typically need\nvast training data before they can be put to any good use. The data generation\nprocess for dynamical systems can be an expensive endeavor both in terms of\ntime and resources. Active learning addresses this shortcoming by acquiring the\nmost informative data, thereby reducing the need to collect enormous datasets.\nWhat makes the current work unique is integrating the deep active learning\nframework into nonlinear system identification. We formulate a general static\ndeep active learning acquisition problem for nonlinear system identification.\nThis is enabled by exploring system dynamics locally in different regions of\nthe input space to obtain a simulated dataset covering the broader input space.\nThis simulated dataset can be used in a static deep active learning acquisition\nscheme referred to as global explorations. The global exploration acquires a\nbatch of initial states corresponding to the most informative state-action\ntrajectories according to a batch acquisition function. The local exploration\nsolves an optimal control problem, finding the control trajectory that\nmaximizes some measure of information. After a batch of informative initial\nstates is acquired, a new round of local explorations from the initial states\nin the batch is conducted to obtain a set of corresponding control trajectories\nthat are to be applied on the system dynamics to get data from the system.\nInformation measures used in the acquisition scheme are derived from the\npredictive variance of an ensemble of neural networks. The novel method\noutperforms standard data acquisition methods used for system identification of\nnonlinear dynamical systems in the case study performed on simulated data.\n", "date": "2023-02-24", "categories": []}
{"id": "2302.12018", "title": "Gaussian Switch Sampling: A Second Order Approach to Active Learning", "abstract": "  In active learning, acquisition functions define informativeness directly on\nthe representation position within the model manifold. However, for most\nmachine learning models (in particular neural networks) this representation is\nnot fixed due to the training pool fluctuations in between active learning\nrounds. Therefore, several popular strategies are sensitive to experiment\nparameters (e.g. architecture) and do not consider model robustness to\nout-of-distribution settings. To alleviate this issue, we propose a grounded\nsecond-order definition of information content and sample importance within the\ncontext of active learning. Specifically, we define importance by how often a\nneural network \"forgets\" a sample during training - artifacts of second order\nrepresentation shifts. We show that our definition produces highly accurate\nimportance scores even when the model representations are constrained by the\nlack of training data. Motivated by our analysis, we develop Gaussian Switch\nSampling (GauSS). We show that GauSS is setup agnostic and robust to anomalous\ndistributions with exhaustive experiments on three in-distribution benchmarks,\nthree out-of-distribution benchmarks, and three different architectures. We\nreport an improvement of up to 5% when compared against four popular query\nstrategies.\n", "date": "2023-02-16", "categories": []}
{"id": "2211.06224", "title": "ALANNO: An Active Learning Annotation System for Mortals", "abstract": "  Supervised machine learning has become the cornerstone of today's data-driven\nsociety, increasing the need for labeled data. However, the process of\nacquiring labels is often expensive and tedious. One possible remedy is to use\nactive learning (AL) -- a special family of machine learning algorithms\ndesigned to reduce labeling costs. Although AL has been successful in practice,\na number of practical challenges hinder its effectiveness and are often\noverlooked in existing AL annotation tools. To address these challenges, we\ndeveloped ALANNO, an open-source annotation system for NLP tasks equipped with\nfeatures to make AL effective in real-world annotation projects. ALANNO\nfacilitates annotation management in a multi-annotator setup and supports a\nvariety of AL methods and underlying models, which are easily configurable and\nextensible.\n", "date": "2022-11-11", "categories": []}
{"id": "2302.10185", "title": "Active Learning in Brain Tumor Segmentation with Uncertainty Sampling,\n  Annotation Redundancy Restriction, and Data Initialization", "abstract": "  Deep learning models have demonstrated great potential in medical 3D imaging,\nbut their development is limited by the expensive, large volume of annotated\ndata required. Active learning (AL) addresses this by training a model on a\nsubset of the most informative data samples without compromising performance.\nWe compared different AL strategies and propose a framework that minimizes the\namount of data needed for state-of-the-art performance. 638 multi-institutional\nbrain tumor MRI images were used to train a 3D U-net model and compare AL\nstrategies. We investigated uncertainty sampling, annotation redundancy\nrestriction, and initial dataset selection techniques. Uncertainty estimation\ntechniques including Bayesian estimation with dropout, bootstrapping, and\nmargins sampling were compared to random query. Strategies to avoid annotation\nredundancy by removing similar images within the to-be-annotated subset were\nconsidered as well. We determined the minimum amount of data necessary to\nachieve similar performance to the model trained on the full dataset ({\\alpha}\n= 0.1). A variance-based selection strategy using radiomics to identify the\ninitial training dataset is also proposed. Bayesian approximation with dropout\nat training and testing showed similar results to that of the full data model\nwith less than 20% of the training data (p=0.293) compared to random query\nachieving similar performance at 56.5% of the training data (p=0.814).\nAnnotation redundancy restriction techniques achieved state-of-the-art\nperformance at approximately 40%-50% of the training data. Radiomics dataset\ninitialization had higher Dice with initial dataset sizes of 20 and 80 images,\nbut improvements were not significant. In conclusion, we investigated various\nAL strategies with dropout uncertainty estimation achieving state-of-the-art\nperformance with the least annotated data.\n", "date": "2023-02-05", "categories": []}
{"id": "2302.10679", "title": "Evaluating the effect of data augmentation and BALD heuristics on\n  distillation of Semantic-KITTI dataset", "abstract": "  Active Learning (AL) has remained relatively unexplored for LiDAR perception\ntasks in autonomous driving datasets. In this study we evaluate Bayesian active\nlearning methods applied to the task of dataset distillation or core subset\nselection (subset with near equivalent performance as full dataset). We also\nstudy the effect of application of data augmentation (DA) within Bayesian AL\nbased dataset distillation. We perform these experiments on the full\nSemantic-KITTI dataset. We extend our study over our existing work only on\n1/4th of the same dataset. Addition of DA and BALD have a negative impact over\nthe labeling efficiency and thus the capacity to distill datasets. We\ndemonstrate key issues in designing a functional AL framework and finally\nconclude with a review of challenges in real world active learning.\n", "date": "2023-02-21", "categories": []}
{"id": "2301.05490", "title": "Scalable Batch Acquisition for Deep Bayesian Active Learning", "abstract": "  In deep active learning, it is especially important to choose multiple\nexamples to markup at each step to work efficiently, especially on large\ndatasets. At the same time, existing solutions to this problem in the Bayesian\nsetup, such as BatchBALD, have significant limitations in selecting a large\nnumber of examples, associated with the exponential complexity of computing\nmutual information for joint random variables. We, therefore, present the Large\nBatchBALD algorithm, which gives a well-grounded approximation to the BatchBALD\nmethod that aims to achieve comparable quality while being more computationally\nefficient. We provide a complexity analysis of the algorithm, showing a\nreduction in computation time, especially for large batches. Furthermore, we\npresent an extensive set of experimental results on image and text data, both\non toy datasets and larger ones such as CIFAR-100.\n", "date": "2023-01-13", "categories": []}
{"id": "2302.08436", "title": "Trieste: Efficiently Exploring The Depths of Black-box Functions with\n  TensorFlow", "abstract": "  We present Trieste, an open-source Python package for Bayesian optimization\nand active learning benefiting from the scalability and efficiency of\nTensorFlow. Our library enables the plug-and-play of popular TensorFlow-based\nmodels within sequential decision-making loops, e.g. Gaussian processes from\nGPflow or GPflux, or neural networks from Keras. This modular mindset is\ncentral to the package and extends to our acquisition functions and the\ninternal dynamics of the decision-making loop, both of which can be tailored\nand extended by researchers or engineers when tackling custom use cases.\nTrieste is a research-friendly and production-ready toolkit backed by a\ncomprehensive test suite, extensive documentation, and available at\nhttps://github.com/secondmind-labs/trieste.\n", "date": "2023-02-16", "categories": []}
{"id": "2302.06976", "title": "Investigating Multi-source Active Learning for Natural Language\n  Inference", "abstract": "  In recent years, active learning has been successfully applied to an array of\nNLP tasks. However, prior work often assumes that training and test data are\ndrawn from the same distribution. This is problematic, as in real-life settings\ndata may stem from several sources of varying relevance and quality. We show\nthat four popular active learning schemes fail to outperform random selection\nwhen applied to unlabelled pools comprised of multiple data sources on the task\nof natural language inference. We reveal that uncertainty-based strategies\nperform poorly due to the acquisition of collective outliers, i.e.,\nhard-to-learn instances that hamper learning and generalization. When outliers\nare removed, strategies are found to recover and outperform random baselines.\nIn further analysis, we find that collective outliers vary in form between\nsources, and show that hard-to-learn data is not always categorically harmful.\nLastly, we leverage dataset cartography to introduce difficulty-stratified\ntesting and find that different strategies are affected differently by example\nlearnability and difficulty.\n", "date": "2023-02-14", "categories": []}
{"id": "2210.08205", "title": "Active Learning from the Web", "abstract": "  Labeling data is one of the most costly processes in machine learning\npipelines. Active learning is a standard approach to alleviating this problem.\nPool-based active learning first builds a pool of unlabelled data and\niteratively selects data to be labeled so that the total number of required\nlabels is minimized, keeping the model performance high. Many effective\ncriteria for choosing data from the pool have been proposed in the literature.\nHowever, how to build the pool is less explored. Specifically, most of the\nmethods assume that a task-specific pool is given for free. In this paper, we\nadvocate that such a task-specific pool is not always available and propose the\nuse of a myriad of unlabelled data on the Web for the pool for which active\nlearning is applied. As the pool is extremely large, it is likely that relevant\ndata exist in the pool for many tasks, and we do not need to explicitly design\nand build the pool for each task. The challenge is that we cannot compute the\nacquisition scores of all data exhaustively due to the size of the pool. We\npropose an efficient method, Seafaring, to retrieve informative data in terms\nof active learning from the Web using a user-side information retrieval\nalgorithm. In the experiments, we use the online Flickr environment as the pool\nfor active learning. This pool contains more than ten billion images and is\nseveral orders of magnitude larger than the existing pools in the literature\nfor active learning. We confirm that our method performs better than existing\napproaches of using a small unlabelled pool.\n", "date": "2022-10-15", "categories": []}
{"id": "2210.10109", "title": "A Survey of Active Learning for Natural Language Processing", "abstract": "  In this work, we provide a survey of active learning (AL) for its\napplications in natural language processing (NLP). In addition to a\nfine-grained categorization of query strategies, we also investigate several\nother important aspects of applying AL to NLP problems. These include AL for\nstructured prediction tasks, annotation cost, model learning (especially with\ndeep neural models), and starting and stopping AL. Finally, we conclude with a\ndiscussion of related topics and future directions.\n", "date": "2022-10-18", "categories": []}
{"id": "2301.12052", "title": "Leveraging Importance Weights in Subset Selection", "abstract": "  We present a subset selection algorithm designed to work with arbitrary model\nfamilies in a practical batch setting. In such a setting, an algorithm can\nsample examples one at a time but, in order to limit overhead costs, is only\nable to update its state (i.e. further train model weights) once a large enough\nbatch of examples is selected. Our algorithm, IWeS, selects examples by\nimportance sampling where the sampling probability assigned to each example is\nbased on the entropy of models trained on previously selected batches. IWeS\nadmits significant performance improvement compared to other subset selection\nalgorithms for seven publicly available datasets. Additionally, it is\ncompetitive in an active learning setting, where the label information is not\navailable at selection time. We also provide an initial theoretical analysis to\nsupport our importance weighting approach, proving generalization and sampling\nrate bounds.\n", "date": "2023-01-28", "categories": []}
{"id": "2206.01454", "title": "Indirect Active Learning", "abstract": "  Traditional models of active learning assume a learner can directly\nmanipulate or query a covariate $X$ in order to study its relationship with a\nresponse $Y$. However, if $X$ is a feature of a complex system, it may be\npossible only to indirectly influence $X$ by manipulating a control variable\n$Z$, a scenario we refer to as Indirect Active Learning. Under a nonparametric\nmodel of Indirect Active Learning with a fixed budget, we study minimax\nconvergence rates for estimating the relationship between $X$ and $Y$ locally\nat a point, obtaining different rates depending on the complexities and noise\nlevels of the relationships between $Z$ and $X$ and between $X$ and $Y$. We\nalso identify minimax rates for passive learning under comparable assumptions.\nIn many cases, our results show that, while there is an asymptotic benefit to\nactive learning, this benefit is fully realized by a simple two-stage learner\nthat runs two passive experiments in sequence.\n", "date": "2022-06-03", "categories": []}
{"id": "2301.09490", "title": "Speeding Up BatchBALD: A k-BALD Family of Approximations for Active\n  Learning", "abstract": "  Active learning is a powerful method for training machine learning models\nwith limited labeled data. One commonly used technique for active learning is\nBatchBALD, which uses Bayesian neural networks to find the most informative\npoints to label in a pool set. However, BatchBALD can be very slow to compute,\nespecially for larger datasets. In this paper, we propose a new approximation,\nk-BALD, which uses k-wise mutual information terms to approximate BatchBALD,\nmaking it much less expensive to compute. Results on the MNIST dataset show\nthat k-BALD is significantly faster than BatchBALD while maintaining similar\nperformance. Additionally, we also propose a dynamic approach for choosing k\nbased on the quality of the approximation, making it more efficient for larger\ndatasets.\n", "date": "2023-01-23", "categories": []}
{"id": "2210.00423", "title": "Improved Algorithms for Neural Active Learning", "abstract": "  We improve the theoretical and empirical performance of\nneural-network(NN)-based active learning algorithms for the non-parametric\nstreaming setting. In particular, we introduce two regret metrics by minimizing\nthe population loss that are more suitable in active learning than the one used\nin state-of-the-art (SOTA) related work. Then, the proposed algorithm leverages\nthe powerful representation of NNs for both exploitation and exploration, has\nthe query decision-maker tailored for $k$-class classification problems with\nthe performance guarantee, utilizes the full feedback, and updates parameters\nin a more practical and efficient manner. These careful designs lead to an\ninstance-dependent regret upper bound, roughly improving by a multiplicative\nfactor $O(\\log T)$ and removing the curse of input dimensionality. Furthermore,\nwe show that the algorithm can achieve the same performance as the\nBayes-optimal classifier in the long run under the hard-margin setting in\nclassification problems. In the end, we use extensive experiments to evaluate\nthe proposed algorithm and SOTA baselines, to show the improved empirical\nperformance.\n", "date": "2022-10-02", "categories": []}
{"id": "2301.06624", "title": "TAAL: Test-time Augmentation for Active Learning in Medical Image\n  Segmentation", "abstract": "  Deep learning methods typically depend on the availability of labeled data,\nwhich is expensive and time-consuming to obtain. Active learning addresses such\neffort by prioritizing which samples are best to annotate in order to maximize\nthe performance of the task model. While frameworks for active learning have\nbeen widely explored in the context of classification of natural images, they\nhave been only sparsely used in medical image segmentation. The challenge\nresides in obtaining an uncertainty measure that reveals the best candidate\ndata for annotation. This paper proposes Test-time Augmentation for Active\nLearning (TAAL), a novel semi-supervised active learning approach for\nsegmentation that exploits the uncertainty information offered by data\ntransformations. Our method applies cross-augmentation consistency during\ntraining and inference to both improve model learning in a semi-supervised\nfashion and identify the most relevant unlabeled samples to annotate next. In\naddition, our consistency loss uses a modified version of the JSD to further\nimprove model performance. By relying on data transformations rather than on\nexternal modules or simple heuristics typically used in uncertainty-based\nstrategies, TAAL emerges as a simple, yet powerful task-agnostic\nsemi-supervised active learning approach applicable to the medical domain. Our\nresults on a publicly-available dataset of cardiac images show that TAAL\noutperforms existing baseline methods in both fully-supervised and\nsemi-supervised settings. Our implementation is publicly available on\nhttps://github.com/melinphd/TAAL.\n", "date": "2023-01-16", "categories": []}
{"id": "2210.07805", "title": "Meta-Query-Net: Resolving Purity-Informativeness Dilemma in Open-set\n  Active Learning", "abstract": "  Unlabeled data examples awaiting annotations contain open-set noise\ninevitably. A few active learning studies have attempted to deal with this\nopen-set noise for sample selection by filtering out the noisy examples.\nHowever, because focusing on the purity of examples in a query set leads to\noverlooking the informativeness of the examples, the best balancing of purity\nand informativeness remains an important question. In this paper, to solve this\npurity-informativeness dilemma in open-set active learning, we propose a novel\nMeta-Query-Net,(MQ-Net) that adaptively finds the best balancing between the\ntwo factors. Specifically, by leveraging the multi-round property of active\nlearning, we train MQ-Net using a query set without an additional validation\nset. Furthermore, a clear dominance relationship between unlabeled examples is\neffectively captured by MQ-Net through a novel skyline regularization.\nExtensive experiments on multiple open-set active learning scenarios\ndemonstrate that the proposed MQ-Net achieves 20.14% improvement in terms of\naccuracy, compared with the state-of-the-art methods.\n", "date": "2022-10-13", "categories": []}
{"id": "2301.05106", "title": "Forgetful Active Learning with Switch Events: Efficient Sampling for\n  Out-of-Distribution Data", "abstract": "  This paper considers deep out-of-distribution active learning. In practice,\nfully trained neural networks interact randomly with out-of-distribution (OOD)\ninputs and map aberrant samples randomly within the model representation space.\nSince data representations are direct manifestations of the training\ndistribution, the data selection process plays a crucial role in outlier\nrobustness. For paradigms such as active learning, this is especially\nchallenging since protocols must not only improve performance on the training\ndistribution most effectively but further render a robust representation space.\nHowever, existing strategies directly base the data selection on the data\nrepresentation of the unlabeled data which is random for OOD samples by\ndefinition. For this purpose, we introduce forgetful active learning with\nswitch events (FALSE) - a novel active learning protocol for\nout-of-distribution active learning. Instead of defining sample importance on\nthe data representation directly, we formulate \"informativeness\" with learning\ndifficulty during training. Specifically, we approximate how often the network\n\"forgets\" unlabeled samples and query the most \"forgotten\" samples for\nannotation. We report up to 4.5\\% accuracy improvements in over 270\nexperiments, including four commonly used protocols, two OOD benchmarks, one\nin-distribution benchmark, and three different architectures.\n", "date": "2023-01-12", "categories": []}
{"id": "2301.04420", "title": "Combining Self-labeling with Selective Sampling", "abstract": "  Since data is the fuel that drives machine learning models, and access to\nlabeled data is generally expensive, semi-supervised methods are constantly\npopular. They enable the acquisition of large datasets without the need for too\nmany expert labels. This work combines self-labeling techniques with active\nlearning in a selective sampling scenario. We propose a new method that builds\nan ensemble classifier. Based on an evaluation of the inconsistency of the\ndecisions of the individual base classifiers for a given observation, a\ndecision is made on whether to request a new label or use the self-labeling. In\npreliminary studies, we show that naive application of self-labeling can harm\nperformance by introducing bias towards selected classes and consequently lead\nto skewed class distribution. Hence, we also propose mechanisms to reduce this\nphenomenon. Experimental evaluation shows that the proposed method matches\ncurrent selective sampling methods or achieves better results.\n", "date": "2023-01-11", "categories": []}
{"id": "2301.02909", "title": "How to Allocate your Label Budget? Choosing between Active Learning and\n  Learning to Reject in Anomaly Detection", "abstract": "  Anomaly detection attempts at finding examples that deviate from the expected\nbehaviour. Usually, anomaly detection is tackled from an unsupervised\nperspective because anomalous labels are rare and difficult to acquire.\nHowever, the lack of labels makes the anomaly detector have high uncertainty in\nsome regions, which usually results in poor predictive performance or low user\ntrust in the predictions. One can reduce such uncertainty by collecting\nspecific labels using Active Learning (AL), which targets examples close to the\ndetector's decision boundary. Alternatively, one can increase the user trust by\nallowing the detector to abstain from making highly uncertain predictions,\nwhich is called Learning to Reject (LR). One way to do this is by thresholding\nthe detector's uncertainty based on where its performance is low, which\nrequires labels to be evaluated. Although both AL and LR need labels, they work\nwith different types of labels: AL seeks strategic labels, which are evidently\nbiased, while LR requires i.i.d. labels to evaluate the detector's performance\nand set the rejection threshold. Because one usually has a unique label budget,\ndeciding how to optimally allocate it is challenging. In this paper, we propose\na mixed strategy that, given a budget of labels, decides in multiple rounds\nwhether to use the budget to collect AL labels or LR labels. The strategy is\nbased on a reward function that measures the expected gain when allocating the\nbudget to either side. We evaluate our strategy on 18 benchmark datasets and\ncompare it to some baselines.\n", "date": "2023-01-07", "categories": []}
{"id": "2301.03252", "title": "Active Learning for Abstractive Text Summarization", "abstract": "  Construction of human-curated annotated datasets for abstractive text\nsummarization (ATS) is very time-consuming and expensive because creating each\ninstance requires a human annotator to read a long document and compose a\nshorter summary that would preserve the key information relayed by the original\ndocument. Active Learning (AL) is a technique developed to reduce the amount of\nannotation required to achieve a certain level of machine learning model\nperformance. In information extraction and text classification, AL can reduce\nthe amount of labor up to multiple times. Despite its potential for aiding\nexpensive annotation, as far as we know, there were no effective AL query\nstrategies for ATS. This stems from the fact that many AL strategies rely on\nuncertainty estimation, while as we show in our work, uncertain instances are\nusually noisy, and selecting them can degrade the model performance compared to\npassive annotation. We address this problem by proposing the first effective\nquery strategy for AL in ATS based on diversity principles. We show that given\na certain annotation budget, using our strategy in AL annotation helps to\nimprove the model performance in terms of ROUGE and consistency scores.\nAdditionally, we analyze the effect of self-learning and show that it can\nfurther increase the performance of the model.\n", "date": "2023-01-09", "categories": []}
{"id": "2301.01531", "title": "MoBYv2AL: Self-supervised Active Learning for Image Classification", "abstract": "  Active learning(AL) has recently gained popularity for deep learning(DL)\nmodels. This is due to efficient and informative sampling, especially when the\nlearner requires large-scale labelled datasets. Commonly, the sampling and\ntraining happen in stages while more batches are added. One main bottleneck in\nthis strategy is the narrow representation learned by the model that affects\nthe overall AL selection.\n  We present MoBYv2AL, a novel self-supervised active learning framework for\nimage classification. Our contribution lies in lifting MoBY, one of the most\nsuccessful self-supervised learning algorithms, to the AL pipeline. Thus, we\nadd the downstream task-aware objective function and optimize it jointly with\ncontrastive loss. Further, we derive a data-distribution selection function\nfrom labelling the new examples. Finally, we test and study our pipeline\nrobustness and performance for image classification tasks. We successfully\nachieved state-of-the-art results when compared to recent AL methods. Code\navailable: https://github.com/razvancaramalau/MoBYv2AL\n", "date": "2023-01-04", "categories": []}
{"id": "2212.14658", "title": "Deep Active Learning Using Barlow Twins", "abstract": "  The generalisation performance of a convolutional neural networks (CNN) is\nmajorly predisposed by the quantity, quality, and diversity of the training\nimages. All the training data needs to be annotated in-hand before, in many\nreal-world applications data is easy to acquire but expensive and\ntime-consuming to label. The goal of the Active learning for the task is to\ndraw most informative samples from the unlabeled pool which can used for\ntraining after annotation. With total different objective, self-supervised\nlearning which have been gaining meteoric popularity by closing the gap in\nperformance with supervised methods on large computer vision benchmarks.\nself-supervised learning (SSL) these days have shown to produce low-level\nrepresentations that are invariant to distortions of the input sample and can\nencode invariance to artificially created distortions, e.g. rotation,\nsolarization, cropping etc. self-supervised learning (SSL) approaches rely on\nsimpler and more scalable frameworks for learning. In this paper, we unify\nthese two families of approaches from the angle of active learning using\nself-supervised learning mainfold and propose Deep Active Learning using\nBarlowTwins(DALBT), an active learning method for all the datasets using\ncombination of classifier trained along with self-supervised loss framework of\nBarlow Twins to a setting where the model can encode the invariance of\nartificially created distortions, e.g. rotation, solarization, cropping etc.\n", "date": "2022-12-30", "categories": []}
{"id": "2212.13974", "title": "Adversarial Virtual Exemplar Learning for Label-Frugal Satellite Image\n  Change Detection", "abstract": "  Satellite image change detection aims at finding occurrences of targeted\nchanges in a given scene taken at different instants. This task is highly\nchallenging due to the acquisition conditions and also to the subjectivity of\nchanges. In this paper, we investigate satellite image change detection using\nactive learning. Our method is interactive and relies on a question and answer\nmodel which asks the oracle (user) questions about the most informative display\n(dubbed as virtual exemplars), and according to the user's responses, updates\nchange detections. The main contribution of our method consists in a novel\nadversarial model that allows frugally probing the oracle with only the most\nrepresentative, diverse and uncertain virtual exemplars. The latter are learned\nto challenge the most the trained change decision criteria which ultimately\nleads to a better re-estimate of these criteria in the following iterations of\nactive learning. Conducted experiments show the out-performance of our proposed\nadversarial display model against other display strategies as well as the\nrelated work.\n", "date": "2022-12-28", "categories": []}
{"id": "2211.14819", "title": "Deep Active Learning for Computer Vision: Past and Future", "abstract": "  As an important data selection schema, active learning emerges as the\nessential component when iterating an Artificial Intelligence (AI) model. It\nbecomes even more critical given the dominance of deep neural network based\nmodels, which are composed of a large number of parameters and data hungry, in\napplication. Despite its indispensable role for developing AI models, research\non active learning is not as intensive as other research directions. In this\npaper, we present a review of active learning through deep active learning\napproaches from the following perspectives: 1) technical advancements in active\nlearning, 2) applications of active learning in computer vision, 3) industrial\nsystems leveraging or with potential to leverage active learning for data\niteration, 4) current limitations and future research directions. We expect\nthis paper to clarify the significance of active learning in a modern AI model\nmanufacturing process and to bring additional research attention to active\nlearning. By addressing data automation challenges and coping with automated\nmachine learning systems, active learning will facilitate democratization of AI\ntechnologies by boosting model production at scale.\n", "date": "2022-11-27", "categories": []}
{"id": "2212.11360", "title": "Feature Acquisition using Monte Carlo Tree Search", "abstract": "  Feature acquisition algorithms address the problem of acquiring informative\nfeatures while balancing the costs of acquisition to improve the learning\nperformances of ML models. Previous approaches have focused on calculating the\nexpected utility values of features to determine the acquisition sequences.\nOther approaches formulated the problem as a Markov Decision Process (MDP) and\napplied reinforcement learning based algorithms. In comparison to previous\napproaches, we focus on 1) formulating the feature acquisition problem as a MDP\nand applying Monte Carlo Tree Search, 2) calculating the intermediary rewards\nfor each acquisition step based on model improvements and acquisition costs and\n3) simultaneously optimizing model improvement and acquisition costs with\nmulti-objective Monte Carlo Tree Search. With Proximal Policy Optimization and\nDeep Q-Network algorithms as benchmark, we show the effectiveness of our\nproposed approach with experimental study.\n", "date": "2022-12-21", "categories": []}
{"id": "2212.10613", "title": "Temporal Output Discrepancy for Loss Estimation-based Active Learning", "abstract": "  While deep learning succeeds in a wide range of tasks, it highly depends on\nthe massive collection of annotated data which is expensive and time-consuming.\nTo lower the cost of data annotation, active learning has been proposed to\ninteractively query an oracle to annotate a small proportion of informative\nsamples in an unlabeled dataset. Inspired by the fact that the samples with\nhigher loss are usually more informative to the model than the samples with\nlower loss, in this paper we present a novel deep active learning approach that\nqueries the oracle for data annotation when the unlabeled sample is believed to\nincorporate high loss. The core of our approach is a measurement Temporal\nOutput Discrepancy (TOD) that estimates the sample loss by evaluating the\ndiscrepancy of outputs given by models at different optimization steps. Our\ntheoretical investigation shows that TOD lower-bounds the accumulated sample\nloss thus it can be used to select informative unlabeled samples. On basis of\nTOD, we further develop an effective unlabeled data sampling strategy as well\nas an unsupervised learning criterion for active learning. Due to the\nsimplicity of TOD, our methods are efficient, flexible, and task-agnostic.\nExtensive experimental results demonstrate that our approach achieves superior\nperformances than the state-of-the-art active learning methods on image\nclassification and semantic segmentation tasks. In addition, we show that TOD\ncan be utilized to select the best model of potentially the highest testing\naccuracy from a pool of candidate models.\n", "date": "2022-12-20", "categories": []}
{"id": "2212.10836", "title": "Towards Rapid Prototyping and Comparability in Active Learning for Deep\n  Object Detection", "abstract": "  Active learning as a paradigm in deep learning is especially important in\napplications involving intricate perception tasks such as object detection\nwhere labels are difficult and expensive to acquire. Development of active\nlearning methods in such fields is highly computationally expensive and time\nconsuming which obstructs the progression of research and leads to a lack of\ncomparability between methods. In this work, we propose and investigate a\nsandbox setup for rapid development and transparent evaluation of active\nlearning in deep object detection. Our experiments with commonly used\nconfigurations of datasets and detection architectures found in the literature\nshow that results obtained in our sandbox environment are representative of\nresults on standard configurations. The total compute time to obtain results\nand assess the learning behavior can thereby be reduced by factors of up to 14\nwhen comparing with Pascal VOC and up to 32 when comparing with BDD100k. This\nallows for testing and evaluating data acquisition and labeling strategies in\nunder half a day and contributes to the transparency and development speed in\nthe field of active learning for object detection.\n", "date": "2022-12-21", "categories": []}
{"id": "2212.11109", "title": "MAViC: Multimodal Active Learning for Video Captioning", "abstract": "  A large number of annotated video-caption pairs are required for training\nvideo captioning models, resulting in high annotation costs. Active learning\ncan be instrumental in reducing these annotation requirements. However, active\nlearning for video captioning is challenging because multiple semantically\nsimilar captions are valid for a video, resulting in high entropy outputs even\nfor less-informative samples. Moreover, video captioning algorithms are\nmultimodal in nature with a visual encoder and language decoder. Further, the\nsequential and combinatorial nature of the output makes the problem even more\nchallenging. In this paper, we introduce MAViC which leverages our proposed\nMultimodal Semantics Aware Sequential Entropy (M-SASE) based acquisition\nfunction to address the challenges of active learning approaches for video\ncaptioning. Our approach integrates semantic similarity and uncertainty of both\nvisual and language dimensions in the acquisition function. Our detailed\nexperiments empirically demonstrate the efficacy of M-SASE for active learning\nfor video captioning and improve on the baselines by a large margin.\n", "date": "2022-12-11", "categories": []}
{"id": "2212.09510", "title": "Near-optimal Policy Identification in Active Reinforcement Learning", "abstract": "  Many real-world reinforcement learning tasks require control of complex\ndynamical systems that involve both costly data acquisition processes and large\nstate spaces. In cases where the transition dynamics can be readily evaluated\nat specified states (e.g., via a simulator), agents can operate in what is\noften referred to as planning with a \\emph{generative model}. We propose the\nAE-LSVI algorithm for best-policy identification, a novel variant of the\nkernelized least-squares value iteration (LSVI) algorithm that combines\noptimism with pessimism for active exploration (AE). AE-LSVI provably\nidentifies a near-optimal policy \\emph{uniformly} over an entire state space\nand achieves polynomial sample complexity guarantees that are independent of\nthe number of states. When specialized to the recently introduced offline\ncontextual Bayesian optimization setting, our algorithm achieves improved\nsample complexity bounds. Experimentally, we demonstrate that AE-LSVI\noutperforms other RL algorithms in a variety of environments when robustness to\nthe initial state is required.\n", "date": "2022-12-19", "categories": []}
{"id": "2212.07568", "title": "Man-recon: manifold learning for reconstruction with deep autoencoder\n  for smart seismic interpretation", "abstract": "  Deep learning can extract rich data representations if provided sufficient\nquantities of labeled training data. For many tasks however, annotating data\nhas significant costs in terms of time and money, owing to the high standards\nof subject matter expertise required, for example in medical and geophysical\nimage interpretation tasks. Active Learning can identify the most informative\ntraining examples for the interpreter to train, leading to higher efficiency.\nWe propose an Active learning method based on jointly learning representations\nfor supervised and unsupervised tasks. The learned manifold structure is later\nutilized to identify informative training samples most dissimilar from the\nlearned manifold from the error profiles on the unsupervised task. We verify\nthe efficiency of the proposed method on a seismic facies segmentation dataset\nfrom the Netherlands F3 block survey, significantly outperforming contemporary\nmethods to achieve the highest mean Intersection-Over-Union value of 0.773.\n", "date": "2022-12-15", "categories": []}
{"id": "2212.06364", "title": "ALRt: An Active Learning Framework for Irregularly Sampled Temporal Data", "abstract": "  Sepsis is a deadly condition affecting many patients in the hospital. Recent\nstudies have shown that patients diagnosed with sepsis have significant\nmortality and morbidity, resulting from the body's dysfunctional host response\nto infection. Clinicians often rely on the use of Sequential Organ Failure\nAssessment (SOFA), Systemic Inflammatory Response Syndrome (SIRS), and the\nModified Early Warning Score (MEWS) to identify early signs of clinical\ndeterioration requiring further work-up and treatment. However, many of these\ntools are manually computed and were not designed for automated computation.\nThere have been different methods used for developing sepsis onset models, but\nmany of these models must be trained on a sufficient number of patient\nobservations in order to form accurate sepsis predictions. Additionally, the\naccurate annotation of patients with sepsis is a major ongoing challenge. In\nthis paper, we propose the use of Active Learning Recurrent Neural Networks\n(ALRts) for short temporal horizons to improve the prediction of irregularly\nsampled temporal events such as sepsis. We show that an active learning RNN\nmodel trained on limited data can form robust sepsis predictions comparable to\nmodels using the entire training dataset.\n", "date": "2022-12-13", "categories": []}
{"id": "2212.06493", "title": "Pixel is All You Need: Adversarial Trajectory-Ensemble Active Learning\n  for Salient Object Detection", "abstract": "  Although weakly-supervised techniques can reduce the labeling effort, it is\nunclear whether a saliency model trained with weakly-supervised data (e.g.,\npoint annotation) can achieve the equivalent performance of its\nfully-supervised version. This paper attempts to answer this unexplored\nquestion by proving a hypothesis: there is a point-labeled dataset where\nsaliency models trained on it can achieve equivalent performance when trained\non the densely annotated dataset. To prove this conjecture, we proposed a novel\nyet effective adversarial trajectory-ensemble active learning (ATAL). Our\ncontributions are three-fold: 1) Our proposed adversarial attack triggering\nuncertainty can conquer the overconfidence of existing active learning methods\nand accurately locate these uncertain pixels. {2)} Our proposed\ntrajectory-ensemble uncertainty estimation method maintains the advantages of\nthe ensemble networks while significantly reducing the computational cost. {3)}\nOur proposed relationship-aware diversity sampling algorithm can conquer\noversampling while boosting performance. Experimental results show that our\nATAL can find such a point-labeled dataset, where a saliency model trained on\nit obtained $97\\%$ -- $99\\%$ performance of its fully-supervised version with\nonly ten annotated points per image.\n", "date": "2022-12-13", "categories": []}
{"id": "2212.04868", "title": "Frugal Reinforcement-based Active Learning", "abstract": "  Most of the existing learning models, particularly deep neural networks, are\nreliant on large datasets whose hand-labeling is expensive and time demanding.\nA current trend is to make the learning of these models frugal and less\ndependent on large collections of labeled data. Among the existing solutions,\ndeep active learning is currently witnessing a major interest and its purpose\nis to train deep networks using as few labeled samples as possible. However,\nthe success of active learning is highly dependent on how critical are these\nsamples when training models. In this paper, we devise a novel active learning\napproach for label-efficient training. The proposed method is iterative and\naims at minimizing a constrained objective function that mixes diversity,\nrepresentativity and uncertainty criteria. The proposed approach is\nprobabilistic and unifies all these criteria in a single objective function\nwhose solution models the probability of relevance of samples (i.e., how\ncritical) when learning a decision function. We also introduce a novel\nweighting mechanism based on reinforcement learning, which adaptively balances\nthese criteria at each training iteration, using a particular stateless\nQ-learning model. Extensive experiments conducted on staple image\nclassification data, including Object-DOTA, show the effectiveness of our\nproposed model w.r.t. several baselines including random, uncertainty and flat\nas well as other work.\n", "date": "2022-12-09", "categories": []}
{"id": "2212.04211", "title": "Evaluating Zero-cost Active Learning for Object Detection", "abstract": "  Object detection requires substantial labeling effort for learning robust\nmodels. Active learning can reduce this effort by intelligently selecting\nrelevant examples to be annotated. However, selecting these examples properly\nwithout introducing a sampling bias with a negative impact on the\ngeneralization performance is not straightforward and most active learning\ntechniques can not hold their promises on real-world benchmarks. In our\nevaluation paper, we focus on active learning techniques without a\ncomputational overhead besides inference, something we refer to as zero-cost\nactive learning. In particular, we show that a key ingredient is not only the\nscore on a bounding box level but also the technique used for aggregating the\nscores for ranking images. We outline our experimental setup and also discuss\npractical considerations when using active learning for object detection.\n", "date": "2022-12-08", "categories": []}
{"id": "2212.03088", "title": "An Empirical Study on the Efficacy of Deep Active Learning for Image\n  Classification", "abstract": "  Deep Active Learning (DAL) has been advocated as a promising method to reduce\nlabeling costs in supervised learning. However, existing evaluations of DAL\nmethods are based on different settings, and their results are controversial.\nTo tackle this issue, this paper comprehensively evaluates 19 existing DAL\nmethods in a uniform setting, including traditional\nfully-\\underline{s}upervised \\underline{a}ctive \\underline{l}earning (SAL)\nstrategies and emerging \\underline{s}emi-\\underline{s}upervised\n\\underline{a}ctive \\underline{l}earning (SSAL) techniques. We have several\nnon-trivial findings. First, most SAL methods cannot achieve higher accuracy\nthan random selection. Second, semi-supervised training brings significant\nperformance improvement compared to pure SAL methods. Third, performing data\nselection in the SSAL setting can achieve a significant and consistent\nperformance improvement, especially with abundant unlabeled data. Our findings\nproduce the following guidance for practitioners: one should (i) apply SSAL\nearly and (ii) collect more unlabeled data whenever possible, for better model\nperformance.\n", "date": "2022-11-30", "categories": []}
{"id": "2212.01703", "title": "Active learning using adaptable task-based prioritisation", "abstract": "  Supervised machine learning-based medical image computing applications\nnecessitate expert label curation, while unlabelled image data might be\nrelatively abundant. Active learning methods aim to prioritise a subset of\navailable image data for expert annotation, for label-efficient model training.\nWe develop a controller neural network that measures priority of images in a\nsequence of batches, as in batch-mode active learning, for multi-class\nsegmentation tasks. The controller is optimised by rewarding positive\ntask-specific performance gain, within a Markov decision process (MDP)\nenvironment that also optimises the task predictor. In this work, the task\npredictor is a segmentation network. A meta-reinforcement learning algorithm is\nproposed with multiple MDPs, such that the pre-trained controller can be\nadapted to a new MDP that contains data from different institutes and/or\nrequires segmentation of different organs or structures within the abdomen. We\npresent experimental results using multiple CT datasets from more than one\nthousand patients, with segmentation tasks of nine different abdominal organs,\nto demonstrate the efficacy of the learnt prioritisation controller function\nand its cross-institute and cross-organ adaptability. We show that the proposed\nadaptable prioritisation metric yields converging segmentation accuracy for the\nnovel class of kidney, unseen in training, using between approximately 40\\% to\n60\\% of labels otherwise required with other heuristic or random prioritisation\nmetrics. For clinical datasets of limited size, the proposed adaptable\nprioritisation offers a performance improvement of 22.6\\% and 10.2\\% in Dice\nscore, for tasks of kidney and liver vessel segmentation, respectively,\ncompared to random prioritisation and alternative active sampling strategies.\n", "date": "2022-12-03", "categories": []}
{"id": "2212.01968", "title": "Dissimilar Nodes Improve Graph Active Learning", "abstract": "  Training labels for graph embedding algorithms could be costly to obtain in\nmany practical scenarios. Active learning (AL) algorithms are very helpful to\nobtain the most useful labels for training while keeping the total number of\nlabel queries under a certain budget. The existing Active Graph Embedding\nframework proposes to use centrality score, density score, and entropy score to\nevaluate the value of unlabeled nodes, and it has been shown to be capable of\nbringing some improvement to the node classification tasks of Graph\nConvolutional Networks. However, when evaluating the importance of unlabeled\nnodes, it fails to consider the influence of existing labeled nodes on the\nvalue of unlabeled nodes. In other words, given the same unlabeled node, the\ncomputed informative score is always the same and is agnostic to the labeled\nnode set. With the aim to address this limitation, in this work, we introduce 3\ndissimilarity-based information scores for active learning: feature\ndissimilarity score (FDS), structure dissimilarity score (SDS), and embedding\ndissimilarity score (EDS). We find out that those three scores are able to take\nthe influence of the labeled set on the value of unlabeled candidates into\nconsideration, boosting our AL performance. According to experiments, our newly\nproposed scores boost the classification accuracy by 2.1% on average and are\ncapable of generalizing to different Graph Neural Network architectures.\n", "date": "2022-12-05", "categories": []}
{"id": "2212.00965", "title": "AL-iGAN: An Active Learning Framework for Tunnel Geological\n  Reconstruction Based on TBM Operational Data", "abstract": "  In tunnel boring machine (TBM) underground projects, an accurate description\nof the rock-soil types distributed in the tunnel can decrease the construction\nrisk ({\\it e.g.} surface settlement and landslide) and improve the efficiency\nof construction. In this paper, we propose an active learning framework, called\nAL-iGAN, for tunnel geological reconstruction based on TBM operational data.\nThis framework contains two main parts: one is the usage of active learning\ntechniques for recommending new drilling locations to label the TBM operational\ndata and then to form new training samples; and the other is an incremental\ngenerative adversarial network for geological reconstruction (iGAN-GR), whose\nweights can be incrementally updated to improve the reconstruction performance\nby using the new samples. The numerical experiment validate the effectiveness\nof the proposed framework as well.\n", "date": "2022-12-02", "categories": []}
{"id": "2208.14856", "title": "Active Learning with Effective Scoring Functions for Semi-Supervised\n  Temporal Action Localization", "abstract": "  Temporal Action Localization (TAL) aims to predict both action category and\ntemporal boundary of action instances in untrimmed videos, i.e., start and end\ntime. Fully-supervised solutions are usually adopted in most existing works,\nand proven to be effective. One of the practical bottlenecks in these solutions\nis the large amount of labeled training data required. To reduce expensive\nhuman label cost, this paper focuses on a rarely investigated yet practical\ntask named semi-supervised TAL and proposes an effective active learning\nmethod, named AL-STAL. We leverage four steps for actively selecting video\nsamples with high informativeness and training the localization model, named\n\\emph{Train, Query, Annotate, Append}. Two scoring functions that consider the\nuncertainty of localization model are equipped in AL-STAL, thus facilitating\nthe video sample rank and selection. One takes entropy of predicted label\ndistribution as measure of uncertainty, named Temporal Proposal Entropy (TPE).\nAnd the other introduces a new metric based on mutual information between\nadjacent action proposals and evaluates the informativeness of video samples,\nnamed Temporal Context Inconsistency (TCI). To validate the effectiveness of\nproposed method, we conduct extensive experiments on two benchmark datasets\nTHUMOS'14 and ActivityNet 1.3. Experiment results show that AL-STAL outperforms\nthe existing competitors and achieves satisfying performance compared with\nfully-supervised learning.\n", "date": "2022-08-31", "categories": []}
{"id": "2211.16942", "title": "ALARM: Active LeArning of Rowhammer Mitigations", "abstract": "  Rowhammer is a serious security problem of contemporary dynamic random-access\nmemory (DRAM) where reads or writes of bits can flip other bits. DRAM\nmanufacturers add mitigations, but don't disclose details, making it difficult\nfor customers to evaluate their efficacy. We present a tool, based on active\nlearning, that automatically infers parameter of Rowhammer mitigations against\nsynthetic models of modern DRAM.\n", "date": "2022-11-30", "categories": []}
{"id": "2207.13339", "title": "ALBench: A Framework for Evaluating Active Learning in Object Detection", "abstract": "  Active learning is an important technology for automated machine learning\nsystems. In contrast to Neural Architecture Search (NAS) which aims at\nautomating neural network architecture design, active learning aims at\nautomating training data selection. It is especially critical for training a\nlong-tailed task, in which positive samples are sparsely distributed. Active\nlearning alleviates the expensive data annotation issue through incrementally\ntraining models powered with efficient data selection. Instead of annotating\nall unlabeled samples, it iteratively selects and annotates the most valuable\nsamples. Active learning has been popular in image classification, but has not\nbeen fully explored in object detection. Most of current approaches on object\ndetection are evaluated with different settings, making it difficult to fairly\ncompare their performance. To facilitate the research in this field, this paper\ncontributes an active learning benchmark framework named as ALBench for\nevaluating active learning in object detection. Developed on an automatic deep\nmodel training system, this ALBench framework is easy-to-use, compatible with\ndifferent active learning algorithms, and ensures the same training and testing\nprotocols. We hope this automated benchmark system help researchers to easily\nreproduce literature's performance and have objective comparisons with prior\narts. The code will be release through Github.\n", "date": "2022-07-27", "categories": []}
{"id": "2211.13587", "title": "Responsible Active Learning via Human-in-the-loop Peer Study", "abstract": "  Active learning has been proposed to reduce data annotation efforts by only\nmanually labelling representative data samples for training. Meanwhile, recent\nactive learning applications have benefited a lot from cloud computing services\nwith not only sufficient computational resources but also crowdsourcing\nframeworks that include many humans in the active learning loop. However,\nprevious active learning methods that always require passing large-scale\nunlabelled data to cloud may potentially raise significant data privacy issues.\nTo mitigate such a risk, we propose a responsible active learning method,\nnamely Peer Study Learning (PSL), to simultaneously preserve data privacy and\nimprove model stability. Specifically, we first introduce a human-in-the-loop\nteacher-student architecture to isolate unlabelled data from the task learner\n(teacher) on the cloud-side by maintaining an active learner (student) on the\nclient-side. During training, the task learner instructs the light-weight\nactive learner which then provides feedback on the active sampling criterion.\nTo further enhance the active learner via large-scale unlabelled data, we\nintroduce multiple peer students into the active learner which is trained by a\nnovel learning paradigm, including the In-Class Peer Study on labelled data and\nthe Out-of-Class Peer Study on unlabelled data. Lastly, we devise a\ndiscrepancy-based active sampling criterion, Peer Study Feedback, that exploits\nthe variability of peer students to select the most informative data to improve\nmodel stability. Extensive experiments demonstrate the superiority of the\nproposed PSL over a wide range of active learning methods in both standard and\nsensitive protection settings.\n", "date": "2022-11-24", "categories": []}
{"id": "2211.13786", "title": "PyTAIL: Interactive and Incremental Learning of NLP Models with Human in\n  the Loop for Online Data", "abstract": "  Online data streams make training machine learning models hard because of\ndistribution shift and new patterns emerging over time. For natural language\nprocessing (NLP) tasks that utilize a collection of features based on lexicons\nand rules, it is important to adapt these features to the changing data. To\naddress this challenge we introduce PyTAIL, a python library, which allows a\nhuman in the loop approach to actively train NLP models. PyTAIL enhances\ngeneric active learning, which only suggests new instances to label by also\nsuggesting new features like rules and lexicons to label. Furthermore, PyTAIL\nis flexible enough for users to accept, reject, or update rules and lexicons as\nthe model is being trained. Finally, we simulate the performance of PyTAIL on\nexisting social media benchmark datasets for text classification. We compare\nvarious active learning strategies on these benchmarks. The model closes the\ngap with as few as 10% of the training data. Finally, we also highlight the\nimportance of tracking evaluation metric on remaining data (which is not yet\nmerged with active learning) alongside the test dataset. This highlights the\neffectiveness of the model in accurately annotating the remaining dataset,\nwhich is especially suitable for batch processing of large unlabelled corpora.\nPyTAIL will be available at https://github.com/socialmediaie/pytail.\n", "date": "2022-11-24", "categories": []}
{"id": "2211.11546", "title": "PartAL: Efficient Partial Active Learning in Multi-Task Visual Settings", "abstract": "  Multi-task learning is central to many real-world applications.\nUnfortunately, obtaining labelled data for all tasks is time-consuming,\nchallenging, and expensive. Active Learning (AL) can be used to reduce this\nburden. Existing techniques typically involve picking images to be annotated\nand providing annotations for all tasks.\n  In this paper, we show that it is more effective to select not only the\nimages to be annotated but also a subset of tasks for which to provide\nannotations at each AL iteration. Furthermore, the annotations that are\nprovided can be used to guess pseudo-labels for the tasks that remain\nunannotated. We demonstrate the effectiveness of our approach on several\npopular multi-task datasets.\n", "date": "2022-11-21", "categories": []}
{"id": "2211.10013", "title": "Active Learning by Query by Committee with Robust Divergences", "abstract": "  Active learning is a widely used methodology for various problems with high\nmeasurement costs. In active learning, the next object to be measured is\nselected by an acquisition function, and measurements are performed\nsequentially. The query by committee is a well-known acquisition function. In\nconventional methods, committee disagreement is quantified by the\nKullback--Leibler divergence. In this paper, the measure of disagreement is\ndefined by the Bregman divergence, which includes the Kullback--Leibler\ndivergence as an instance, and the dual $\\gamma$-power divergence. As a\nparticular class of the Bregman divergence, the $\\beta$-divergence is\nconsidered. By deriving the influence function, we show that the proposed\nmethod using $\\beta$-divergence and dual $\\gamma$-power divergence are more\nrobust than the conventional method in which the measure of disagreement is\ndefined by the Kullback--Leibler divergence. Experimental results show that the\nproposed method performs as well as or better than the conventional method.\n", "date": "2022-11-18", "categories": []}
{"id": "2209.10729", "title": "Fair Robust Active Learning by Joint Inconsistency", "abstract": "  Fairness and robustness play vital roles in trustworthy machine learning.\nObserving safety-critical needs in various annotation-expensive vision\napplications, we introduce a novel learning framework, Fair Robust Active\nLearning (FRAL), generalizing conventional active learning to fair and\nadversarial robust scenarios. This framework allows us to achieve standard and\nrobust minimax fairness with limited acquired labels. In FRAL, we then observe\nexisting fairness-aware data selection strategies suffer from either\nineffectiveness under severe data imbalance or inefficiency due to huge\ncomputations of adversarial training. To address these two problems, we develop\na novel Joint INconsistency (JIN) method exploiting prediction inconsistencies\nbetween benign and adversarial inputs as well as between standard and robust\nmodels. These two inconsistencies can be used to identify potential fairness\ngains and data imbalance mitigations. Thus, by performing label acquisition\nwith our inconsistency-based ranking metrics, we can alleviate the class\nimbalance issue and enhance minimax fairness with limited computation.\nExtensive experiments on diverse datasets and sensitive groups demonstrate that\nour method obtains the best results in standard and robust fairness under\nwhite-box PGD attacks compared with existing active data selection baselines.\n", "date": "2022-09-22", "categories": []}
{"id": "2211.09283", "title": "Active Learning with Expected Error Reduction", "abstract": "  Active learning has been studied extensively as a method for efficient data\ncollection. Among the many approaches in literature, Expected Error Reduction\n(EER) (Roy and McCallum) has been shown to be an effective method for active\nlearning: select the candidate sample that, in expectation, maximally decreases\nthe error on an unlabeled set. However, EER requires the model to be retrained\nfor every candidate sample and thus has not been widely used for modern deep\nneural networks due to this large computational cost. In this paper we\nreformulate EER under the lens of Bayesian active learning and derive a\ncomputationally efficient version that can use any Bayesian parameter sampling\nmethod (such as arXiv:1506.02142). We then compare the empirical performance of\nour method using Monte Carlo dropout for parameter sampling against state of\nthe art methods in the deep active learning literature. Experiments are\nperformed on four standard benchmark datasets and three WILDS datasets\n(arXiv:2012.07421). The results indicate that our method outperforms all other\nmethods except one in the data shift scenario: a model dependent,\nnon-information theoretic method that requires an order of magnitude higher\ncomputational cost (arXiv:1906.03671).\n", "date": "2022-11-17", "categories": []}
{"id": "2211.08735", "title": "Can Strategic Data Collection Improve the Performance of Poverty\n  Prediction Models?", "abstract": "  Machine learning-based estimates of poverty and wealth are increasingly being\nused to guide the targeting of humanitarian aid and the allocation of social\nassistance. However, the ground truth labels used to train these models are\ntypically borrowed from existing surveys that were designed to produce national\nstatistics -- not to train machine learning models. Here, we test whether\nadaptive sampling strategies for ground truth data collection can improve the\nperformance of poverty prediction models. Through simulations, we compare the\nstatus quo sampling strategies (uniform at random and stratified random\nsampling) to alternatives that prioritize acquiring training data based on\nmodel uncertainty or model performance on sub-populations. Perhaps\nsurprisingly, we find that none of these active learning methods improve over\nuniform-at-random sampling. We discuss how these results can help shape future\nefforts to refine machine learning-based estimates of poverty.\n", "date": "2022-11-16", "categories": []}
{"id": "2211.08112", "title": "An Efficient Active Learning Pipeline for Legal Text Classification", "abstract": "  Active Learning (AL) is a powerful tool for learning with less labeled data,\nin particular, for specialized domains, like legal documents, where unlabeled\ndata is abundant, but the annotation requires domain expertise and is thus\nexpensive. Recent works have shown the effectiveness of AL strategies for\npre-trained language models. However, most AL strategies require a set of\nlabeled samples to start with, which is expensive to acquire. In addition,\npre-trained language models have been shown unstable during fine-tuning with\nsmall datasets, and their embeddings are not semantically meaningful. In this\nwork, we propose a pipeline for effectively using active learning with\npre-trained language models in the legal domain. To this end, we leverage the\navailable unlabeled data in three phases. First, we continue pre-training the\nmodel to adapt it to the downstream task. Second, we use knowledge distillation\nto guide the model's embeddings to a semantically meaningful space. Finally, we\npropose a simple, yet effective, strategy to find the initial set of labeled\nsamples with fewer actions compared to existing methods. Our experiments on\nContract-NLI, adapted to the classification task, and LEDGAR benchmarks show\nthat our approach outperforms standard AL strategies, and is more efficient.\nFurthermore, our pipeline reaches comparable results to the fully-supervised\napproach with a small performance gap, and dramatically reduced annotation\ncost. Code and the adapted data will be made available.\n", "date": "2022-11-15", "categories": []}
{"id": "2211.05997", "title": "LiDAL: Inter-frame Uncertainty Based Active Learning for 3D LiDAR\n  Semantic Segmentation", "abstract": "  We propose LiDAL, a novel active learning method for 3D LiDAR semantic\nsegmentation by exploiting inter-frame uncertainty among LiDAR frames. Our core\nidea is that a well-trained model should generate robust results irrespective\nof viewpoints for scene scanning and thus the inconsistencies in model\npredictions across frames provide a very reliable measure of uncertainty for\nactive sample selection. To implement this uncertainty measure, we introduce\nnew inter-frame divergence and entropy formulations, which serve as the metrics\nfor active selection. Moreover, we demonstrate additional performance gains by\npredicting and incorporating pseudo-labels, which are also selected using the\nproposed inter-frame uncertainty measure. Experimental results validate the\neffectiveness of LiDAL: we achieve 95% of the performance of fully supervised\nlearning with less than 5% of annotations on the SemanticKITTI and nuScenes\ndatasets, outperforming state-of-the-art active learning methods. Code release:\nhttps://github.com/hzykent/LiDAL.\n", "date": "2022-11-11", "categories": []}
{"id": "2211.04128", "title": "Active Learning with Tabular Language Models", "abstract": "  Despite recent advancements in tabular language model research, real-world\napplications are still challenging. In industry, there is an abundance of\ntables found in spreadsheets, but acquisition of substantial amounts of labels\nis expensive, since only experts can annotate the often highly technical and\ndomain-specific tables. Active learning could potentially reduce labeling\ncosts, however, so far there are no works related to active learning in\nconjunction with tabular language models. In this paper we investigate\ndifferent acquisition functions in a real-world industrial tabular language\nmodel use case for sub-cell named entity recognition. Our results show that\ncell-level acquisition functions with built-in diversity can significantly\nreduce the labeling effort, while enforced table diversity is detrimental. We\nfurther see open fundamental questions concerning computational efficiency and\nthe perspective of human annotators.\n", "date": "2022-11-08", "categories": []}
{"id": "2208.00549", "title": "Unifying Approaches in Active Learning and Active Sampling via Fisher\n  Information and Information-Theoretic Quantities", "abstract": "  Recently proposed methods in data subset selection, that is active learning\nand active sampling, use Fisher information, Hessians, similarity matrices\nbased on gradients, and gradient lengths to estimate how informative data is\nfor a model's training. Are these different approaches connected, and if so,\nhow? We revisit the fundamentals of Bayesian optimal experiment design and show\nthat these recently proposed methods can be understood as approximations to\ninformation-theoretic quantities: among them, the mutual information between\npredictions and model parameters, known as expected information gain or BALD in\nmachine learning, and the mutual information between predictions of acquisition\ncandidates and test samples, known as expected predictive information gain. We\ndevelop a comprehensive set of approximations using Fisher information and\nobserved information and derive a unified framework that connects seemingly\ndisparate literature. Although Bayesian methods are often seen as separate from\nnon-Bayesian ones, the sometimes fuzzy notion of \"informativeness\" expressed in\nvarious non-Bayesian objectives leads to the same couple of information\nquantities, which were, in principle, already known by Lindley (1956) and\nMacKay (1992).\n", "date": "2022-08-01", "categories": []}
{"id": "2211.00246", "title": "Batch Active Learning from the Perspective of Sparse Approximation", "abstract": "  Active learning enables efficient model training by leveraging interactions\nbetween machine learning agents and human annotators. We study and propose a\nnovel framework that formulates batch active learning from the sparse\napproximation's perspective. Our active learning method aims to find an\ninformative subset from the unlabeled data pool such that the corresponding\ntraining loss function approximates its full data pool counterpart. We realize\nthe framework as sparsity-constrained discontinuous optimization problems,\nwhich explicitly balance uncertainty and representation for large-scale\napplications and could be solved by greedy or proximal iterative hard\nthresholding algorithms. The proposed method can adapt to various settings,\nincluding both Bayesian and non-Bayesian neural networks. Numerical experiments\nshow that our work achieves competitive performance across different settings\nwith lower computational complexity.\n", "date": "2022-11-01", "categories": []}
{"id": "2211.02233", "title": "Improved Adaptive Algorithm for Scalable Active Learning with Weak\n  Labeler", "abstract": "  Active learning with strong and weak labelers considers a practical setting\nwhere we have access to both costly but accurate strong labelers and inaccurate\nbut cheap predictions provided by weak labelers. We study this problem in the\nstreaming setting, where decisions must be taken \\textit{online}. We design a\nnovel algorithmic template, Weak Labeler Active Cover (WL-AC), that is able to\nrobustly leverage the lower quality weak labelers to reduce the query\ncomplexity while retaining the desired level of accuracy. Prior active learning\nalgorithms with access to weak labelers learn a difference classifier which\npredicts where the weak labels differ from strong labelers; this requires the\nstrong assumption of realizability of the difference classifier (Zhang and\nChaudhuri,2015). WL-AC bypasses this \\textit{realizability} assumption and thus\nis applicable to many real-world scenarios such as random corrupted weak labels\nand high dimensional family of difference classifiers (\\textit{e.g.,} deep\nneural nets). Moreover, WL-AC cleverly trades off evaluating the quality with\nfull exploitation of weak labelers, which allows to convert any active learning\nstrategy to one that can leverage weak labelers. We provide an instantiation of\nthis template that achieves the optimal query complexity for any given weak\nlabeler, without knowing its accuracy a-priori. Empirically, we propose an\ninstantiation of the WL-AC template that can be efficiently implemented for\nlarge-scale models (\\textit{e.g}., deep neural nets) and show its effectiveness\non the corrupted-MNIST dataset by significantly reducing the number of labels\nwhile keeping the same accuracy as in passive learning.\n", "date": "2022-11-04", "categories": []}
{"id": "2211.01354", "title": "Improving Named Entity Recognition in Telephone Conversations via\n  Effective Active Learning with Human in the Loop", "abstract": "  Telephone transcription data can be very noisy due to speech recognition\nerrors, disfluencies, etc. Not only that annotating such data is very\nchallenging for the annotators, but also such data may have lots of annotation\nerrors even after the annotation job is completed, resulting in a very poor\nmodel performance. In this paper, we present an active learning framework that\nleverages human in the loop learning to identify data samples from the\nannotated dataset for re-annotation that are more likely to contain annotation\nerrors. In this way, we largely reduce the need for data re-annotation for the\nwhole dataset. We conduct extensive experiments with our proposed approach for\nNamed Entity Recognition and observe that by re-annotating only about 6%\ntraining instances out of the whole dataset, the F1 score for a certain entity\ntype can be significantly improved by about 25%.\n", "date": "2022-11-02", "categories": []}
{"id": "2208.05379", "title": "Multi-task Active Learning for Pre-trained Transformer-based Models", "abstract": "  Multi-task learning, in which several tasks are jointly learned by a single\nmodel, allows NLP models to share information from multiple annotations and may\nfacilitate better predictions when the tasks are inter-related. This technique,\nhowever, requires annotating the same text with multiple annotation schemes\nwhich may be costly and laborious. Active learning (AL) has been demonstrated\nto optimize annotation processes by iteratively selecting unlabeled examples\nwhose annotation is most valuable for the NLP model. Yet, multi-task active\nlearning (MT-AL) has not been applied to state-of-the-art pre-trained\nTransformer-based NLP models. This paper aims to close this gap. We explore\nvarious multi-task selection criteria in three realistic multi-task scenarios,\nreflecting different relations between the participating tasks, and demonstrate\nthe effectiveness of multi-task compared to single-task selection. Our results\nsuggest that MT-AL can be effectively used in order to minimize annotation\nefforts for multi-task NLP models.\n", "date": "2022-08-10", "categories": []}
{"id": "2210.15696", "title": "COMET-QE and Active Learning for Low-Resource Machine Translation", "abstract": "  Active learning aims to deliver maximum benefit when resources are scarce. We\nuse COMET-QE, a reference-free evaluation metric, to select sentences for\nlow-resource neural machine translation. Using Swahili, Kinyarwanda and Spanish\nfor our experiments, we show that COMET-QE significantly outperforms two\nvariants of Round Trip Translation Likelihood (RTTL) and random sentence\nselection by up to 5 BLEU points for 20k sentences selected by Active Learning\non a 30k baseline. This suggests that COMET-QE is a powerful tool for sentence\nselection in the very low-resource limit.\n", "date": "2022-10-27", "categories": []}
{"id": "2210.15786", "title": "Poisson Reweighted Laplacian Uncertainty Sampling for Graph-based Active\n  Learning", "abstract": "  We show that uncertainty sampling is sufficient to achieve exploration versus\nexploitation in graph-based active learning, as long as the measure of\nuncertainty properly aligns with the underlying model and the model properly\nreflects uncertainty in unexplored regions. In particular, we use a recently\ndeveloped algorithm, Poisson ReWeighted Laplace Learning (PWLL) for the\nclassifier and we introduce an acquisition function designed to measure\nuncertainty in this graph-based classifier that identifies unexplored regions\nof the data. We introduce a diagonal perturbation in PWLL which produces\nexponential localization of solutions, and controls the exploration versus\nexploitation tradeoff in active learning. We use the well-posed continuum limit\nof PWLL to rigorously analyze our method, and present experimental results on a\nnumber of graph-based image classification problems.\n", "date": "2022-10-27", "categories": []}
{"id": "2210.14576", "title": "Uncertainty Sentence Sampling by Virtual Adversarial Perturbation", "abstract": "  Active learning for sentence understanding attempts to reduce the annotation\ncost by identifying the most informative examples. Common methods for active\nlearning use either uncertainty or diversity sampling in the pool-based\nscenario. In this work, to incorporate both predictive uncertainty and sample\ndiversity, we propose Virtual Adversarial Perturbation for Active Learning\n(VAPAL) , an uncertainty-diversity combination framework, using virtual\nadversarial perturbation (Miyato et al., 2019) as model uncertainty\nrepresentation. VAPAL consistently performs equally well or even better than\nthe strong baselines on four sentence understanding datasets: AGNEWS, IMDB,\nPUBMED, and SST-2, offering a potential option for active learning on sentence\nunderstanding tasks.\n", "date": "2022-10-26", "categories": []}
{"id": "2210.14465", "title": "Eeny, meeny, miny, moe. How to choose data for morphological inflection", "abstract": "  Data scarcity is a widespread problem in numerous natural language processing\n(NLP) tasks for low-resource languages. Within morphology, the labour-intensive\nwork of tagging/glossing data is a serious bottleneck for both NLP and language\ndocumentation. Active learning (AL) aims to reduce the cost of data annotation\nby selecting data that is most informative for improving the model. In this\npaper, we explore four sampling strategies for the task of morphological\ninflection using a Transformer model: a pair of oracle experiments where data\nis chosen based on whether the model already can or cannot inflect the test\nforms correctly, as well as strategies based on high/low model confidence,\nentropy, as well as random selection. We investigate the robustness of each\nstrategy across 30 typologically diverse languages. We also perform a more\nin-depth case study of Nat\\\"ugu. Our results show a clear benefit to selecting\ndata based on model confidence and entropy. Unsurprisingly, the oracle\nexperiment, where only incorrectly handled forms are chosen for further\ntraining, which is presented as a proxy for linguist/language consultant\nfeedback, shows the most improvement. This is followed closely by choosing\nlow-confidence and high-entropy predictions. We also show that despite the\nconventional wisdom of larger data sets yielding better accuracy, introducing\nmore instances of high-confidence or low-entropy forms, or forms that the model\ncan already inflect correctly, can reduce model performance.\n", "date": "2022-10-26", "categories": []}
{"id": "2210.12704", "title": "Batch Multi-Fidelity Active Learning with Budget Constraints", "abstract": "  Learning functions with high-dimensional outputs is critical in many\napplications, such as physical simulation and engineering design. However,\ncollecting training examples for these applications is often costly, e.g. by\nrunning numerical solvers. The recent work (Li et al., 2022) proposes the first\nmulti-fidelity active learning approach for high-dimensional outputs, which can\nacquire examples at different fidelities to reduce the cost while improving the\nlearning performance. However, this method only queries at one pair of fidelity\nand input at a time, and hence has a risk to bring in strongly correlated\nexamples to reduce the learning efficiency. In this paper, we propose Batch\nMulti-Fidelity Active Learning with Budget Constraints (BMFAL-BC), which can\npromote the diversity of training examples to improve the benefit-cost ratio,\nwhile respecting a given budget constraint for batch queries. Hence, our method\ncan be more practically useful. Specifically, we propose a novel batch\nacquisition function that measures the mutual information between a batch of\nmulti-fidelity queries and the target function, so as to penalize highly\ncorrelated queries and encourages diversity. The optimization of the batch\nacquisition function is challenging in that it involves a combinatorial search\nover many fidelities while subject to the budget constraint. To address this\nchallenge, we develop a weighted greedy algorithm that can sequentially\nidentify each (fidelity, input) pair, while achieving a near $(1 -\n1/e)$-approximation of the optimum. We show the advantage of our method in\nseveral computational physics and engineering applications.\n", "date": "2022-10-23", "categories": []}
{"id": "2210.12122", "title": "Targeted active learning for probabilistic models", "abstract": "  A fundamental task in science is to design experiments that yield valuable\ninsights about the system under study. Mathematically, these insights can be\nrepresented as a utility or risk function that shapes the value of conducting\neach experiment. We present PDBAL, a targeted active learning method that\nadaptively designs experiments to maximize scientific utility. PDBAL takes a\nuser-specified risk function and combines it with a probabilistic model of the\nexperimental outcomes to choose designs that rapidly converge on a high-utility\nmodel. We prove theoretical bounds on the label complexity of PDBAL and provide\nfast closed-form solutions for designing experiments with common exponential\nfamily likelihoods. In simulation studies, PDBAL consistently outperforms\nstandard untargeted approaches that focus on maximizing expected information\ngain over the design space. Finally, we demonstrate the scientific potential of\nPDBAL through a study on a large cancer drug screen dataset where PDBAL quickly\nrecovers the most efficacious drugs with a small fraction of the total number\nof experiments.\n", "date": "2022-10-21", "categories": []}
{"id": "2210.10586", "title": "Active Learning for Imbalanced Civil Infrastructure Data", "abstract": "  Aging civil infrastructures are closely monitored by engineers for damage and\ncritical defects. As the manual inspection of such large structures is costly\nand time-consuming, we are working towards fully automating the visual\ninspections to support the prioritization of maintenance activities. To that\nend we combine recent advances in drone technology and deep learning.\nUnfortunately, annotation costs are incredibly high as our proprietary civil\nengineering dataset must be annotated by highly trained engineers. Active\nlearning is, therefore, a valuable tool to optimize the trade-off between model\nperformance and annotation costs. Our use-case differs from the classical\nactive learning setting as our dataset suffers from heavy class imbalance and\nconsists of a much larger already labeled data pool than other active learning\nresearch. We present a novel method capable of operating in this challenging\nsetting by replacing the traditional active learning acquisition function with\nan auxiliary binary discriminator. We experimentally show that our novel method\noutperforms the best-performing traditional active learning method (BALD) by 5%\nand 38% accuracy on CIFAR-10 and our proprietary dataset respectively.\n", "date": "2022-10-19", "categories": []}
{"id": "2207.02182", "title": "ST-CoNAL: Consistency-Based Acquisition Criterion Using Temporal\n  Self-Ensemble for Active Learning", "abstract": "  Modern deep learning has achieved great success in various fields. However,\nit requires the labeling of huge amounts of data, which is expensive and\nlabor-intensive. Active learning (AL), which identifies the most informative\nsamples to be labeled, is becoming increasingly important to maximize the\nefficiency of the training process. The existing AL methods mostly use only a\nsingle final fixed model for acquiring the samples to be labeled. This strategy\nmay not be good enough in that the structural uncertainty of a model for given\ntraining data is not considered to acquire the samples. In this study, we\npropose a novel acquisition criterion based on temporal self-ensemble generated\nby conventional stochastic gradient descent (SGD) optimization. These\nself-ensemble models are obtained by capturing the intermediate network weights\nobtained through SGD iterations. Our acquisition function relies on a\nconsistency measure between the student and teacher models. The student models\nare given a fixed number of temporal self-ensemble models, and the teacher\nmodel is constructed by averaging the weights of the student models. Using the\nproposed acquisition criterion, we present an AL algorithm, namely\nstudent-teacher consistency-based AL (ST-CoNAL). Experiments conducted for\nimage classification tasks on CIFAR-10, CIFAR-100, Caltech-256, and Tiny\nImageNet datasets demonstrate that the proposed ST-CoNAL achieves significantly\nbetter performance than the existing acquisition methods. Furthermore,\nextensive experiments show the robustness and effectiveness of our methods.\n", "date": "2022-07-05", "categories": []}
{"id": "2210.08403", "title": "Semantic Segmentation with Active Semi-Supervised Representation\n  Learning", "abstract": "  Obtaining human per-pixel labels for semantic segmentation is incredibly\nlaborious, often making labeled dataset construction prohibitively expensive.\nHere, we endeavor to overcome this problem with a novel algorithm that combines\nsemi-supervised and active learning, resulting in the ability to train an\neffective semantic segmentation algorithm with significantly lesser labeled\ndata. To do this, we extend the prior state-of-the-art S4AL algorithm by\nreplacing its mean teacher approach for semi-supervised learning with a\nself-training approach that improves learning with noisy labels. We further\nboost the neural network's ability to query useful data by adding a contrastive\nlearning head, which leads to better understanding of the objects in the scene,\nand hence, better queries for active learning. We evaluate our method on CamVid\nand CityScapes datasets, the de-facto standards for active learning for\nsemantic segmentation. We achieve more than 95% of the network's performance on\nCamVid and CityScapes datasets, utilizing only 12.1% and 15.1% of the labeled\ndata, respectively. We also benchmark our method across existing stand-alone\nsemi-supervised learning methods on the CityScapes dataset and achieve superior\nperformance without any bells or whistles.\n", "date": "2022-10-16", "categories": []}
{"id": "2210.04137", "title": "Few-Shot Continual Active Learning by a Robot", "abstract": "  In this paper, we consider a challenging but realistic continual learning\n(CL) problem, Few-Shot Continual Active Learning (FoCAL), where a CL agent is\nprovided with unlabeled data for a new or a previously learned task in each\nincrement and the agent only has limited labeling budget available. Towards\nthis, we build on the continual learning and active learning literature and\ndevelop a framework that can allow a CL agent to continually learn new object\nclasses from a few labeled training examples. Our framework represents each\nobject class using a uniform Gaussian mixture model (GMM) and uses\npseudo-rehearsal to mitigate catastrophic forgetting. The framework also uses\nuncertainty measures on the Gaussian representations of the previously learned\nclasses to find the most informative samples to be labeled in an increment. We\nevaluate our approach on the CORe-50 dataset and on a real humanoid robot for\nthe object classification task. The results show that our approach not only\nproduces state-of-the-art results on the dataset but also allows a real robot\nto continually learn unseen objects in a real environment with limited labeling\nsupervision provided by its user.\n", "date": "2022-10-09", "categories": []}
{"id": "2210.06526", "title": "Microscopy is All You Need", "abstract": "  We pose that microscopy offers an ideal real-world experimental environment\nfor the development and deployment of active Bayesian and reinforcement\nlearning methods. Indeed, the tremendous progress achieved by machine learning\n(ML) and artificial intelligence over the last decade has been largely achieved\nvia the utilization of static data sets, from the paradigmatic MNIST to the\nbespoke corpora of text and image data used to train large models such as GPT3,\nDALLE and others. However, it is now recognized that continuous, minute\nimprovements to state-of-the-art do not necessarily translate to advances in\nreal-world applications. We argue that a promising pathway for the development\nof ML methods is via the route of domain-specific deployable algorithms in\nareas such as electron and scanning probe microscopy and chemical imaging. This\nwill benefit both fundamental physical studies and serve as a test bed for more\ncomplex autonomous systems such as robotics and manufacturing. Favorable\nenvironment characteristics of scanning and electron microscopy include low\nrisk, extensive availability of domain-specific priors and rewards, relatively\nsmall effects of exogeneous variables, and often the presence of both upstream\nfirst principles as well as downstream learnable physical models for both\nstatics and dynamics. Recent developments in programmable interfaces, edge\ncomputing, and access to APIs facilitating microscope control, all render the\ndeployment of ML codes on operational microscopes straightforward. We discuss\nthese considerations and hope that these arguments will lead to creating a\nnovel set of development targets for the ML community by accelerating both\nreal-world ML applications and scientific progress.\n", "date": "2022-10-12", "categories": []}
{"id": "2210.06749", "title": "Reducing Annotation Effort by Identifying and Labeling Contextually\n  Diverse Classes for Semantic Segmentation Under Domain Shift", "abstract": "  In Active Domain Adaptation (ADA), one uses Active Learning (AL) to select a\nsubset of images from the target domain, which are then annotated and used for\nsupervised domain adaptation (DA). Given the large performance gap between\nsupervised and unsupervised DA techniques, ADA allows for an excellent\ntrade-off between annotation cost and performance. Prior art makes use of\nmeasures of uncertainty or disagreement of models to identify `regions' to be\nannotated by the human oracle. However, these regions frequently comprise of\npixels at object boundaries which are hard and tedious to annotate. Hence, even\nif the fraction of image pixels annotated reduces, the overall annotation time\nand the resulting cost still remain high. In this work, we propose an ADA\nstrategy, which given a frame, identifies a set of classes that are hardest for\nthe model to predict accurately, thereby recommending semantically meaningful\nregions to be annotated in a selected frame. We show that these set of `hard'\nclasses are context-dependent and typically vary across frames, and when\nannotated help the model generalize better. We propose two ADA techniques: the\nAnchor-based and Augmentation-based approaches to select complementary and\ndiverse regions in the context of the current training set. Our approach\nachieves 66.6 mIoU on GTA to Cityscapes dataset with an annotation budget of\n4.7% in comparison to 64.9 mIoU by MADA using 5% of annotations. Our technique\ncan also be used as a decorator for any existing frame-based AL technique,\ne.g., we report 1.5% performance improvement for CDAL on Cityscapes using our\napproach.\n", "date": "2022-10-13", "categories": []}
{"id": "2210.05770", "title": "Deep Active Ensemble Sampling For Image Classification", "abstract": "  Conventional active learning (AL) frameworks aim to reduce the cost of data\nannotation by actively requesting the labeling for the most informative data\npoints. However, introducing AL to data hungry deep learning algorithms has\nbeen a challenge. Some proposed approaches include uncertainty-based\ntechniques, geometric methods, implicit combination of uncertainty-based and\ngeometric approaches, and more recently, frameworks based on semi/self\nsupervised techniques. In this paper, we address two specific problems in this\narea. The first is the need for efficient exploitation/exploration trade-off in\nsample selection in AL. For this, we present an innovative integration of\nrecent progress in both uncertainty-based and geometric frameworks to enable an\nefficient exploration/exploitation trade-off in sample selection strategy. To\nthis end, we build on a computationally efficient approximate of Thompson\nsampling with key changes as a posterior estimator for uncertainty\nrepresentation. Our framework provides two advantages: (1) accurate posterior\nestimation, and (2) tune-able trade-off between computational overhead and\nhigher accuracy. The second problem is the need for improved training protocols\nin deep AL. For this, we use ideas from semi/self supervised learning to\npropose a general approach that is independent of the specific AL technique\nbeing used. Taken these together, our framework shows a significant improvement\nover the state-of-the-art, with results that are comparable to the performance\nof supervised-learning under the same setting. We show empirical results of our\nframework, and comparative performance with the state-of-the-art on four\ndatasets, namely, MNIST, CIFAR10, CIFAR100 and ImageNet to establish a new\nbaseline in two different settings.\n", "date": "2022-10-11", "categories": []}
{"id": "2210.06028", "title": "VL4Pose: Active Learning Through Out-Of-Distribution Detection For Pose\n  Estimation", "abstract": "  Advances in computing have enabled widespread access to pose estimation,\ncreating new sources of data streams. Unlike mock set-ups for data collection,\ntapping into these data streams through on-device active learning allows us to\ndirectly sample from the real world to improve the spread of the training\ndistribution. However, on-device computing power is limited, implying that any\ncandidate active learning algorithm should have a low compute footprint while\nalso being reliable. Although multiple algorithms cater to pose estimation,\nthey either use extensive compute to power state-of-the-art results or are not\ncompetitive in low-resource settings. We address this limitation with VL4Pose\n(Visual Likelihood For Pose Estimation), a first principles approach for active\nlearning through out-of-distribution detection. We begin with a simple premise:\npose estimators often predict incoherent poses for out-of-distribution samples.\nHence, can we identify a distribution of poses the model has been trained on,\nto identify incoherent poses the model is unsure of? Our solution involves\nmodelling the pose through a simple parametric Bayesian network trained via\nmaximum likelihood estimation. Therefore, poses incurring a low likelihood\nwithin our framework are out-of-distribution samples making them suitable\ncandidates for annotation. We also observe two useful side-outcomes: VL4Pose\nin-principle yields better uncertainty estimates by unifying joint and pose\nlevel ambiguity, as well as the unintentional but welcome ability of VL4Pose to\nperform pose refinement in limited scenarios. We perform qualitative and\nquantitative experiments on three datasets: MPII, LSP and ICVL, spanning human\nand hand pose estimation. Finally, we note that VL4Pose is simple,\ncomputationally inexpensive and competitive, making it suitable for challenging\ntasks such as on-device active learning.\n", "date": "2022-10-12", "categories": []}
{"id": "2208.05406", "title": "Active Sampling of Multiple Sources for Sequential Estimation", "abstract": "  Consider $K$ processes, each generating a sequence of identical and\nindependent random variables. The probability measures of these processes have\nrandom parameters that must be estimated. Specifically, they share a parameter\n$\\theta$ common to all probability measures. Additionally, each process\n$i\\in\\{1, \\dots, K\\}$ has a private parameter $\\alpha_i$. The objective is to\ndesign an active sampling algorithm for sequentially estimating these\nparameters in order to form reliable estimates for all shared and private\nparameters with the fewest number of samples. This sampling algorithm has three\nkey components: (i)~data-driven sampling decisions, which dynamically over time\nspecifies which of the $K$ processes should be selected for sampling;\n(ii)~stopping time for the process, which specifies when the accumulated data\nis sufficient to form reliable estimates and terminate the sampling process;\nand (iii)~estimators for all shared and private parameters. Owing to the\nsequential estimation being known to be analytically intractable, this paper\nadopts \\emph {conditional} estimation cost functions, leading to a sequential\nestimation approach that was recently shown to render tractable analysis.\nAsymptotically optimal decision rules (sampling, stopping, and estimation) are\ndelineated, and numerical experiments are provided to compare the efficacy and\nquality of the proposed procedure with those of the relevant approaches.\n", "date": "2022-08-10", "categories": []}
{"id": "2210.03822", "title": "Is margin all you need? An extensive empirical study of active learning\n  on tabular data", "abstract": "  Given a labeled training set and a collection of unlabeled data, the goal of\nactive learning (AL) is to identify the best unlabeled points to label. In this\ncomprehensive study, we analyze the performance of a variety of AL algorithms\non deep neural networks trained on 69 real-world tabular classification\ndatasets from the OpenML-CC18 benchmark. We consider different data regimes and\nthe effect of self-supervised model pre-training. Surprisingly, we find that\nthe classical margin sampling technique matches or outperforms all others,\nincluding current state-of-art, in a wide range of experimental settings. To\nresearchers, we hope to encourage rigorous benchmarking against margin, and to\npractitioners facing tabular data labeling constraints that\nhyper-parameter-free margin may often be all they need.\n", "date": "2022-10-07", "categories": []}
{"id": "2210.02442", "title": "Making Your First Choice: To Address Cold Start Problem in Vision Active\n  Learning", "abstract": "  Active learning promises to improve annotation efficiency by iteratively\nselecting the most important data to be annotated first. However, we uncover a\nstriking contradiction to this promise: active learning fails to select data as\nefficiently as random selection at the first few choices. We identify this as\nthe cold start problem in vision active learning, caused by a biased and\noutlier initial query. This paper seeks to address the cold start problem by\nexploiting the three advantages of contrastive learning: (1) no annotation is\nrequired; (2) label diversity is ensured by pseudo-labels to mitigate bias; (3)\ntypical data is determined by contrastive features to reduce outliers.\nExperiments are conducted on CIFAR-10-LT and three medical imaging datasets\n(i.e. Colon Pathology, Abdominal CT, and Blood Cell Microscope). Our initial\nquery not only significantly outperforms existing active querying strategies\nbut also surpasses random selection by a large margin. We foresee our solution\nto the cold start problem as a simple yet strong baseline to choose the initial\nquery for vision active learning. Code is available:\nhttps://github.com/c-liangyu/CSVAL\n", "date": "2022-10-05", "categories": []}
{"id": "2210.01329", "title": "Active Learning for Regression with Aggregated Outputs", "abstract": "  Due to the privacy protection or the difficulty of data collection, we cannot\nobserve individual outputs for each instance, but we can observe aggregated\noutputs that are summed over multiple instances in a set in some real-world\napplications. To reduce the labeling cost for training regression models for\nsuch aggregated data, we propose an active learning method that sequentially\nselects sets to be labeled to improve the predictive performance with fewer\nlabeled sets. For the selection measurement, the proposed method uses the\nmutual information, which quantifies the reduction of the uncertainty of the\nmodel parameters by observing the aggregated output. With Bayesian linear basis\nfunctions for modeling outputs given an input, which include approximated\nGaussian processes and neural networks, we can efficiently calculate the mutual\ninformation in a closed form. With the experiments using various datasets, we\ndemonstrate that the proposed method achieves better predictive performance\nwith fewer labeled sets than existing methods.\n", "date": "2022-10-04", "categories": []}
{"id": "2210.01526", "title": "DIAGNOSE: Avoiding Out-of-distribution Data using Submodular Information\n  Measures", "abstract": "  Avoiding out-of-distribution (OOD) data is critical for training supervised\nmachine learning models in the medical imaging domain. Furthermore, obtaining\nlabeled medical data is difficult and expensive since it requires expert\nannotators like doctors, radiologists, etc. Active learning (AL) is a\nwell-known method to mitigate labeling costs by selecting the most diverse or\nuncertain samples. However, current AL methods do not work well in the medical\nimaging domain with OOD data. We propose Diagnose (avoiDing out-of-dIstribution\ndAta usinG submodular iNfOrmation meaSurEs), a novel active learning framework\nthat can jointly model similarity and dissimilarity, which is crucial in mining\nin-distribution data and avoiding OOD data at the same time. Particularly, we\nuse a small number of data points as exemplars that represent a query set of\nin-distribution data points and a private set of OOD data points. We illustrate\nthe generalizability of our framework by evaluating it on a wide variety of\nreal-world OOD scenarios. Our experiments verify the superiority of Diagnose\nover the state-of-the-art AL methods across multiple domains of medical\nimaging.\n", "date": "2022-10-04", "categories": []}
{"id": "2210.01090", "title": "Nonstationary data stream classification with online active learning and\n  siamese neural networks", "abstract": "  We have witnessed in recent years an ever-growing volume of information\nbecoming available in a streaming manner in various application areas. As a\nresult, there is an emerging need for online learning methods that train\npredictive models on-the-fly. A series of open challenges, however, hinder\ntheir deployment in practice. These are, learning as data arrive in real-time\none-by-one, learning from data with limited ground truth information, learning\nfrom nonstationary data, and learning from severely imbalanced data, while\noccupying a limited amount of memory for data storage. We propose the\nActiSiamese algorithm, which addresses these challenges by combining online\nactive learning, siamese networks, and a multi-queue memory. It develops a new\ndensity-based active learning strategy which considers similarity in the latent\n(rather than the input) space. We conduct an extensive study that compares the\nrole of different active learning budgets and strategies, the performance\nwith/without memory, the performance with/without ensembling, in both synthetic\nand real-world datasets, under different data nonstationarity characteristics\nand class imbalance levels. ActiSiamese outperforms baseline and\nstate-of-the-art algorithms, and is effective under severe imbalance, even only\nwhen a fraction of the arriving instances' labels is available. We publicly\nrelease our code to the community.\n", "date": "2022-10-03", "categories": []}
{"id": "2209.13518", "title": "Graph-Based Active Machine Learning Method for Diverse and Novel\n  Antimicrobial Peptides Generation and Selection", "abstract": "  As antibiotic-resistant bacterial strains are rapidly spreading worldwide,\ninfections caused by these strains are emerging as a global crisis causing the\ndeath of millions of people every year. Antimicrobial Peptides (AMPs) are one\nof the candidates to tackle this problem because of their potential diversity,\nand ability to favorably modulate the host immune response. However,\nlarge-scale screening of new AMP candidates is expensive, time-consuming, and\nnow affordable in developing countries, which need the treatments the most. In\nthis work, we propose a novel active machine learning-based framework that\nstatistically minimizes the number of wet-lab experiments needed to design new\nAMPs, while ensuring a high diversity and novelty of generated AMPs sequences,\nin multi-rounds of wet-lab AMP screening settings. Combining recurrent neural\nnetwork models and a graph-based filter (GraphCC), our proposed approach\ndelivers novel and diverse candidates and demonstrates better performances\naccording to our defined metrics.\n", "date": "2022-09-18", "categories": []}
{"id": "2209.06855", "title": "Data Lifecycle Management in Evolving Input Distributions for\n  Learning-based Aerospace Applications", "abstract": "  As input distributions evolve over a mission lifetime, maintaining\nperformance of learning-based models becomes challenging. This paper presents a\nframework to incrementally retrain a model by selecting a subset of test inputs\nto label, which allows the model to adapt to changing input distributions.\nAlgorithms within this framework are evaluated based on (1) model performance\nthroughout mission lifetime and (2) cumulative costs associated with labeling\nand model retraining. We provide an open-source benchmark of a satellite pose\nestimation model trained on images of a satellite in space and deployed in\nnovel scenarios (e.g., different backgrounds or misbehaving pixels), where\nalgorithms are evaluated on their ability to maintain high performance by\nretraining on a subset of inputs. We also propose a novel algorithm to select a\ndiverse subset of inputs for labeling, by characterizing the information gain\nfrom an input using Bayesian uncertainty quantification and choosing a subset\nthat maximizes collective information gain using concepts from batch active\nlearning. We show that our algorithm outperforms others on the benchmark, e.g.,\nachieves comparable performance to an algorithm that labels 100% of inputs,\nwhile only labeling 50% of inputs, resulting in low costs and high performance\nover the mission lifetime.\n", "date": "2022-09-14", "categories": []}
{"id": "2209.11464", "title": "Smart Active Sampling to enhance Quality Assurance Efficiency", "abstract": "  We propose a new sampling strategy, called smart active sapling, for quality\ninspections outside the production line. Based on the principles of active\nlearning a machine learning model decides which samples are sent to quality\ninspection. On the one hand, this minimizes the production of scrap parts due\nto earlier detection of quality violations. On the other hand, quality\ninspection costs are reduced for smooth operation.\n", "date": "2022-09-23", "categories": []}
{"id": "2209.11481", "title": "Active Few-Shot Classification: a New Paradigm for Data-Scarce Learning\n  Settings", "abstract": "  We consider a novel formulation of the problem of Active Few-Shot\nClassification (AFSC) where the objective is to classify a small, initially\nunlabeled, dataset given a very restrained labeling budget. This problem can be\nseen as a rival paradigm to classical Transductive Few-Shot Classification\n(TFSC), as both these approaches are applicable in similar conditions. We first\npropose a methodology that combines statistical inference, and an original\ntwo-tier active learning strategy that fits well into this framework. We then\nadapt several standard vision benchmarks from the field of TFSC. Our\nexperiments show the potential benefits of AFSC can be substantial, with gains\nin average weighted accuracy of up to 10% compared to state-of-the-art TFSC\nmethods for the same labeling budget. We believe this new paradigm could lead\nto new developments and standards in data-scarce learning settings.\n", "date": "2022-09-23", "categories": []}
{"id": "2209.11629", "title": "From Weakly Supervised Learning to Active Learning", "abstract": "  Applied mathematics and machine computations have raised a lot of hope since\nthe recent success of supervised learning. Many practitioners in industries\nhave been trying to switch from their old paradigms to machine learning.\nInterestingly, those data scientists spend more time scrapping, annotating and\ncleaning data than fine-tuning models. This thesis is motivated by the\nfollowing question: can we derive a more generic framework than the one of\nsupervised learning in order to learn from clutter data?\n  This question is approached through the lens of weakly supervised learning,\nassuming that the bottleneck of data collection lies in annotation. We model\nweak supervision as giving, rather than a unique target, a set of target\ncandidates. We argue that one should look for an ``optimistic'' function that\nmatches most of the observations. This allows us to derive a principle to\ndisambiguate partial labels. We also discuss the advantage to incorporate\nunsupervised learning techniques into our framework, in particular manifold\nregularization approached through diffusion techniques, for which we derived a\nnew algorithm that scales better with input dimension then the baseline method.\n  Finally, we switch from passive to active weakly supervised learning,\nintroducing the ``active labeling'' framework, in which a practitioner can\nquery weak information about chosen data. Among others, we leverage the fact\nthat one does not need full information to access stochastic gradients and\nperform stochastic gradient descent.\n", "date": "2022-09-23", "categories": []}
{"id": "2209.07069", "title": "Active Self-Training for Weakly Supervised 3D Scene Semantic\n  Segmentation", "abstract": "  Since the preparation of labeled data for training semantic segmentation\nnetworks of point clouds is a time-consuming process, weakly supervised\napproaches have been introduced to learn from only a small fraction of data.\nThese methods are typically based on learning with contrastive losses while\nautomatically deriving per-point pseudo-labels from a sparse set of\nuser-annotated labels. In this paper, our key observation is that the selection\nof what samples to annotate is as important as how these samples are used for\ntraining. Thus, we introduce a method for weakly supervised segmentation of 3D\nscenes that combines self-training with active learning. The active learning\nselects points for annotation that likely result in performance improvements to\nthe trained model, while the self-training makes efficient use of the\nuser-provided labels for learning the model. We demonstrate that our approach\nleads to an effective method that provides improvements in scene segmentation\nover previous works and baselines, while requiring only a small number of user\nannotations.\n", "date": "2022-09-15", "categories": []}
{"id": "2209.06285", "title": "Warm Start Active Learning with Proxy Labels \\& Selection via\n  Semi-Supervised Fine-Tuning", "abstract": "  Which volume to annotate next is a challenging problem in building medical\nimaging datasets for deep learning. One of the promising methods to approach\nthis question is active learning (AL). However, AL has been a hard nut to crack\nin terms of which AL algorithm and acquisition functions are most useful for\nwhich datasets. Also, the problem is exacerbated with which volumes to label\nfirst when there is zero labeled data to start with. This is known as the cold\nstart problem in AL. We propose two novel strategies for AL specifically for 3D\nimage segmentation. First, we tackle the cold start problem by proposing a\nproxy task and then utilizing uncertainty generated from the proxy task to rank\nthe unlabeled data to be annotated. Second, we craft a two-stage learning\nframework for each active iteration where the unlabeled data is also used in\nthe second stage as a semi-supervised fine-tuning strategy. We show the promise\nof our approach on two well-known large public datasets from medical\nsegmentation decathlon. The results indicate that the initial selection of data\nand semi-supervised framework both showed significant improvement for several\nAL strategies.\n", "date": "2022-09-13", "categories": []}
{"id": "2209.03996", "title": "Active Learning of Classifiers with Label and Seed Queries", "abstract": "  We study exact active learning of binary and multiclass classifiers with\nmargin. Given an $n$-point set $X \\subset \\mathbb{R}^m$, we want to learn any\nunknown classifier on $X$ whose classes have finite strong convex hull margin,\na new notion extending the SVM margin. In the standard active learning setting,\nwhere only label queries are allowed, learning a classifier with strong convex\nhull margin $\\gamma$ requires in the worst case\n$\\Omega\\big(1+\\frac{1}{\\gamma}\\big)^{(m-1)/2}$ queries. On the other hand,\nusing the more powerful seed queries (a variant of equivalence queries), the\ntarget classifier could be learned in $O(m \\log n)$ queries via Littlestone's\nHalving algorithm; however, Halving is computationally inefficient. In this\nwork we show that, by carefully combining the two types of queries, a binary\nclassifier can be learned in time $\\operatorname{poly}(n+m)$ using only $O(m^2\n\\log n)$ label queries and $O\\big(m \\log \\frac{m}{\\gamma}\\big)$ seed queries;\nthe result extends to $k$-class classifiers at the price of a $k!k^2$\nmultiplicative overhead. Similar results hold when the input points have\nbounded bit complexity, or when only one class has strong convex hull margin\nagainst the rest. We complement the upper bounds by showing that in the worst\ncase any algorithm needs $\\Omega\\big(k m \\log \\frac{1}{\\gamma}\\big)$ seed and\nlabel queries to learn a $k$-class classifier with strong convex hull margin\n$\\gamma$.\n", "date": "2022-09-08", "categories": []}
{"id": "2209.01858", "title": "Consistency-Based Semi-supervised Evidential Active Learning for\n  Diagnostic Radiograph Classification", "abstract": "  Deep learning approaches achieve state-of-the-art performance for classifying\nradiology images, but rely on large labelled datasets that require\nresource-intensive annotation by specialists. Both semi-supervised learning and\nactive learning can be utilised to mitigate this annotation burden. However,\nthere is limited work on combining the advantages of semi-supervised and active\nlearning approaches for multi-label medical image classification. Here, we\nintroduce a novel Consistency-based Semi-supervised Evidential Active Learning\nframework (CSEAL). Specifically, we leverage predictive uncertainty based on\ntheories of evidence and subjective logic to develop an end-to-end integrated\napproach that combines consistency-based semi-supervised learning with\nuncertainty-based active learning. We apply our approach to enhance four\nleading consistency-based semi-supervised learning methods: Pseudo-labelling,\nVirtual Adversarial Training, Mean Teacher and NoTeacher. Extensive evaluations\non multi-label Chest X-Ray classification tasks demonstrate that CSEAL achieves\nsubstantive performance improvements over two leading semi-supervised active\nlearning baselines. Further, a class-wise breakdown of results shows that our\napproach can substantially improve accuracy on rarer abnormalities with fewer\nlabelled samples.\n", "date": "2022-09-05", "categories": []}
{"id": "2209.00514", "title": "Efficient Chemical Space Exploration Using Active Learning Based on\n  Marginalized Graph Kernel: an Application for Predicting the Thermodynamic\n  Properties of Alkanes with Molecular Simulation", "abstract": "  We introduce an explorative active learning (AL) algorithm based on Gaussian\nprocess regression and marginalized graph kernel (GPR-MGK) to explore chemical\nspace with minimum cost. Using high-throughput molecular dynamics simulation to\ngenerate data and graph neural network (GNN) to predict, we constructed an\nactive learning molecular simulation framework for thermodynamic property\nprediction. In specific, targeting 251,728 alkane molecules consisting of 4 to\n19 carbon atoms and their liquid physical properties: densities, heat\ncapacities, and vaporization enthalpies, we use the AL algorithm to select the\nmost informative molecules to represent the chemical space. Validation of\ncomputational and experimental test sets shows that only 313 (0.124\\% of the\ntotal) molecules were sufficient to train an accurate GNN model with $\\rm R^2 >\n0.99$ for computational test sets and $\\rm R^2 > 0.94$ for experimental test\nsets. We highlight two advantages of the presented AL algorithm: compatibility\nwith high-throughput data generation and reliable uncertainty quantification.\n", "date": "2022-09-01", "categories": []}
{"id": "2207.05945", "title": "Online Active Regression", "abstract": "  Active regression considers a linear regression problem where the learner\nreceives a large number of data points but can only observe a small number of\nlabels. Since online algorithms can deal with incremental training data and\ntake advantage of low computational cost, we consider an online extension of\nthe active regression problem: the learner receives data points one by one and\nimmediately decides whether it should collect the corresponding labels. The\ngoal is to efficiently maintain the regression of received data points with a\nsmall budget of label queries. We propose novel algorithms for this problem\nunder $\\ell_p$ loss where $p\\in[1,2]$. To achieve a $(1+\\epsilon)$-approximate\nsolution, our proposed algorithms only require\n$\\tilde{\\mathcal{O}}(\\epsilon^{-1} d \\log(n\\kappa))$ queries of labels, where\n$n$ is the number of data points and $\\kappa$ is a quantity, called the\ncondition number, of the data points. The numerical results verify our\ntheoretical results and show that our methods have comparable performance with\noffline active regression algorithms.\n", "date": "2022-07-13", "categories": []}
{"id": "2208.03020", "title": "Deep Bayesian Active-Learning-to-Rank for Endoscopic Image Data", "abstract": "  Automatic image-based disease severity estimation generally uses discrete\n(i.e., quantized) severity labels. Annotating discrete labels is often\ndifficult due to the images with ambiguous severity. An easier alternative is\nto use relative annotation, which compares the severity level between image\npairs. By using a learning-to-rank framework with relative annotation, we can\ntrain a neural network that estimates rank scores that are relative to severity\nlevels. However, the relative annotation for all possible pairs is prohibitive,\nand therefore, appropriate sample pair selection is mandatory. This paper\nproposes a deep Bayesian active-learning-to-rank, which trains a Bayesian\nconvolutional neural network while automatically selecting appropriate pairs\nfor relative annotation. We confirmed the efficiency of the proposed method\nthrough experiments on endoscopic images of ulcerative colitis. In addition, we\nconfirmed that our method is useful even with the severe class imbalance\nbecause of its ability to select samples from minor classes automatically.\n", "date": "2022-08-05", "categories": []}
{"id": "2208.02104", "title": "Active Learning on a Programmable Photonic Quantum Processor", "abstract": "  Training a quantum machine learning model generally requires a large labeled\ndataset, which incurs high labeling and computational costs. To reduce such\ncosts, a selective training strategy, called active learning (AL), chooses only\na subset of the original dataset to learn while maintaining the trained model's\nperformance. Here, we design and implement two AL-enpowered variational quantum\nclassifiers, to investigate the potential applications and effectiveness of AL\nin quantum machine learning. Firstly, we build a programmable free-space\nphotonic quantum processor, which enables the programmed implementation of\nvarious hybrid quantum-classical computing algorithms. Then, we code the\ndesigned variational quantum classifier with AL into the quantum processor, and\nexecute comparative tests for the classifiers with and without the AL strategy.\nThe results validate the great advantage of AL in quantum machine learning, as\nit saves at most $85\\%$ labeling efforts and $91.6\\%$ percent computational\nefforts compared to the training without AL on a data classification task. Our\nresults inspire AL's further applications in large-scale quantum machine\nlearning to drastically reduce training data and speed up training,\nunderpinning the exploration of practical quantum advantages in quantum physics\nor real-world applications.\n", "date": "2022-08-03", "categories": []}
{"id": "2208.00508", "title": "Deep Active Learning with Budget Annotation", "abstract": "  Digital data collected over the decades and data currently being produced\nwith use of information technology is vastly the unlabeled data or data without\ndescription. The unlabeled data is relatively easy to acquire but expensive to\nlabel even with use of domain experts. Most of the recent works focus on use of\nactive learning with uncertainty metrics measure to address this problem.\nAlthough most uncertainty selection strategies are very effective, they fail to\ntake informativeness of the unlabeled instances into account and are prone to\nquerying outliers. In order to address these challenges we propose an hybrid\napproach of computing both the uncertainty and informativeness of an instance,\nthen automaticaly label the computed instances using budget annotator. To\nreduce the annotation cost, we employ the state-of-the-art pre-trained models\nin order to avoid querying information already contained in those models. Our\nextensive experiments on different sets of datasets demonstrate the efficacy of\nthe proposed approach.\n", "date": "2022-07-31", "categories": []}
{"id": "2207.11493", "title": "Active Pointly-Supervised Instance Segmentation", "abstract": "  The requirement of expensive annotations is a major burden for training a\nwell-performed instance segmentation model. In this paper, we present an\neconomic active learning setting, named active pointly-supervised instance\nsegmentation (APIS), which starts with box-level annotations and iteratively\nsamples a point within the box and asks if it falls on the object. The key of\nAPIS is to find the most desirable points to maximize the segmentation accuracy\nwith limited annotation budgets. We formulate this setting and propose several\nuncertainty-based sampling strategies. The model developed with these\nstrategies yields consistent performance gain on the challenging MS-COCO\ndataset, compared against other learning strategies. The results suggest that\nAPIS, integrating the advantages of active learning and point-based\nsupervision, is an effective learning paradigm for label-efficient instance\nsegmentation.\n", "date": "2022-07-23", "categories": []}
{"id": "2207.12302", "title": "Exploiting Diversity of Unlabeled Data for Label-Efficient\n  Semi-Supervised Active Learning", "abstract": "  The availability of large labeled datasets is the key component for the\nsuccess of deep learning. However, annotating labels on large datasets is\ngenerally time-consuming and expensive. Active learning is a research area that\naddresses the issues of expensive labeling by selecting the most important\nsamples for labeling. Diversity-based sampling algorithms are known as integral\ncomponents of representation-based approaches for active learning. In this\npaper, we introduce a new diversity-based initial dataset selection algorithm\nto select the most informative set of samples for initial labeling in the\nactive learning setting. Self-supervised representation learning is used to\nconsider the diversity of samples in the initial dataset selection algorithm.\nAlso, we propose a novel active learning query strategy, which uses\ndiversity-based sampling on consistency-based embeddings. By considering the\nconsistency information with the diversity in the consistency-based embedding\nscheme, the proposed method could select more informative samples for labeling\nin the semi-supervised learning setting. Comparative experiments show that the\nproposed method achieves compelling results on CIFAR-10 and Caltech-101\ndatasets compared with previous active learning approaches by utilizing the\ndiversity of unlabeled data.\n", "date": "2022-07-25", "categories": []}
{"id": "2206.15204", "title": "Data-Efficient Learning via Minimizing Hyperspherical Energy", "abstract": "  Deep learning on large-scale data is dominant nowadays. The unprecedented\nscale of data has been arguably one of the most important driving forces for\nthe success of deep learning. However, there still exist scenarios where\ncollecting data or labels could be extremely expensive, e.g., medical imaging\nand robotics. To fill up this gap, this paper considers the problem of\ndata-efficient learning from scratch using a small amount of representative\ndata. First, we characterize this problem by active learning on homeomorphic\ntubes of spherical manifolds. This naturally generates feasible hypothesis\nclass. With homologous topological properties, we identify an important\nconnection -- finding tube manifolds is equivalent to minimizing hyperspherical\nenergy (MHE) in physical geometry. Inspired by this connection, we propose a\nMHE-based active learning (MHEAL) algorithm, and provide comprehensive\ntheoretical guarantees for MHEAL, covering convergence and generalization\nanalysis. Finally, we demonstrate the empirical performance of MHEAL in a wide\nrange of applications on data-efficient learning, including deep clustering,\ndistribution matching, version space sampling and deep active learning.\n", "date": "2022-06-30", "categories": []}
{"id": "2206.04838", "title": "In Defense of Core-set: A Density-aware Core-set Selection for Active\n  Learning", "abstract": "  Active learning enables the efficient construction of a labeled dataset by\nlabeling informative samples from an unlabeled dataset. In a real-world active\nlearning scenario, considering the diversity of the selected samples is crucial\nbecause many redundant or highly similar samples exist. Core-set approach is\nthe promising diversity-based method selecting diverse samples based on the\ndistance between samples. However, the approach poorly performs compared to the\nuncertainty-based approaches that select the most difficult samples where\nneural models reveal low confidence. In this work, we analyze the feature space\nthrough the lens of the density and, interestingly, observe that locally sparse\nregions tend to have more informative samples than dense regions. Motivated by\nour analysis, we empower the core-set approach with the density-awareness and\npropose a density-aware core-set (DACS). The strategy is to estimate the\ndensity of the unlabeled samples and select diverse samples mainly from sparse\nregions. To reduce the computational bottlenecks in estimating the density, we\nalso introduce a new density approximation based on locality-sensitive hashing.\nExperimental results clearly demonstrate the efficacy of DACS in both\nclassification and regression tasks and specifically show that DACS can produce\nstate-of-the-art performance in a practical scenario. Since DACS is weakly\ndependent on neural architectures, we present a simple yet effective\ncombination method to show that the existing methods can be beneficially\ncombined with DACS.\n", "date": "2022-06-10", "categories": []}
{"id": "2207.05376", "title": "Optimal Clustering with Noisy Queries via Multi-Armed Bandit", "abstract": "  Motivated by many applications, we study clustering with a faulty oracle. In\nthis problem, there are $n$ items belonging to $k$ unknown clusters, and the\nalgorithm is allowed to ask the oracle whether two items belong to the same\ncluster or not. However, the answer from the oracle is correct only with\nprobability $\\frac{1}{2}+\\frac{\\delta}{2}$. The goal is to recover the hidden\nclusters with minimum number of noisy queries. Previous works have shown that\nthe problem can be solved with $O(\\frac{nk\\log n}{\\delta^2} +\n\\text{poly}(k,\\frac{1}{\\delta}, \\log n))$ queries, while\n$\\Omega(\\frac{nk}{\\delta^2})$ queries is known to be necessary. So, for any\nvalues of $k$ and $\\delta$, there is still a non-trivial gap between upper and\nlower bounds. In this work, we obtain the first matching upper and lower bounds\nfor a wide range of parameters. In particular, a new polynomial time algorithm\nwith $O(\\frac{n(k+\\log n)}{\\delta^2} + \\text{poly}(k,\\frac{1}{\\delta}, \\log\nn))$ queries is proposed. Moreover, we prove a new lower bound of\n$\\Omega(\\frac{n\\log n}{\\delta^2})$, which, combined with the existing\n$\\Omega(\\frac{nk}{\\delta^2})$ bound, matches our upper bound up to an additive\n$\\text{poly}(k,\\frac{1}{\\delta},\\log n)$ term. To obtain the new results, our\nmain ingredient is an interesting connection between our problem and\nmulti-armed bandit, which might provide useful insights for other similar\nproblems.\n", "date": "2022-07-12", "categories": []}
{"id": "2206.13298", "title": "Deep Active Learning for Regression Using $\\epsilon$-weighted Hybrid\n  Query Strategy", "abstract": "  Designing an inexpensive approximate surrogate model that captures the\nsalient features of an expensive high-fidelity behavior is a prevalent approach\nin design optimization. In recent times, Deep Learning (DL) models are being\nused as a promising surrogate computational model for engineering problems.\nHowever, the main challenge in creating a DL-based surrogate is to\nsimulate/label a large number of design points, which is time-consuming for\ncomputationally costly and/or high-dimensional engineering problems. In the\npresent work, we propose a novel sampling technique by combining the active\nlearning (AL) method with DL. We call this method $\\epsilon$-weighted hybrid\nquery strategy ($\\epsilon$-HQS) , which focuses on the evaluation of the\nsurrogate at each learning iteration and provides an estimate of the failure\nprobability of the surrogate in the Design Space. By reusing already collected\ntraining and test data, the learned failure probability guides the next\niteration's sampling process to the region of the high probability of failure.\nDuring the empirical evaluation, better accuracy of the surrogate was observed\nin comparison to other methods of sample selection. We empirically evaluated\nthis method in two different engineering design domains, finite element based\nstatic stress analysis of submarine pressure vessel(computationally costly\nprocess) and second submarine propeller design( high dimensional problem).\nhttps://github.com/vardhah/epsilon_weighted_Hybrid_Query_Strategy\n", "date": "2022-06-24", "categories": []}
{"id": "2207.03934", "title": "Active Learning-based Isolation Forest (ALIF): Enhancing Anomaly\n  Detection in Decision Support Systems", "abstract": "  The detection of anomalous behaviours is an emerging need in many\napplications, particularly in contexts where security and reliability are\ncritical aspects. While the definition of anomaly strictly depends on the\ndomain framework, it is often impractical or too time consuming to obtain a\nfully labelled dataset. The use of unsupervised models to overcome the lack of\nlabels often fails to catch domain specific anomalies as they rely on general\ndefinitions of outlier. This paper suggests a new active learning based\napproach, ALIF, to solve this problem by reducing the number of required labels\nand tuning the detector towards the definition of anomaly provided by the user.\nThe proposed approach is particularly appealing in the presence of a Decision\nSupport System (DSS), a case that is increasingly popular in real-world\nscenarios. While it is common that DSS embedded with anomaly detection\ncapabilities rely on unsupervised models, they don't have a way to improve\ntheir performance: ALIF is able to enhance the capabilities of DSS by\nexploiting the user feedback during common operations. ALIF is a lightweight\nmodification of the popular Isolation Forest that proved superior performances\nwith respect to other state-of-art algorithms in a multitude of real anomaly\ndetection datasets.\n", "date": "2022-07-08", "categories": []}
{"id": "2207.02964", "title": "Mitigating shortage of labeled data using clustering-based active\n  learning with diversity exploration", "abstract": "  In this paper, we proposed a new clustering-based active learning framework,\nnamely Active Learning using a Clustering-based Sampling (ALCS), to address the\nshortage of labeled data. ALCS employs a density-based clustering approach to\nexplore the cluster structure from the data without requiring exhaustive\nparameter tuning. A bi-cluster boundary-based sample query procedure is\nintroduced to improve the learning performance for classifying highly\noverlapped classes. Additionally, we developed an effective diversity\nexploration strategy to address the redundancy among queried samples. Our\nexperimental results justified the efficacy of the ALCS approach.\n", "date": "2022-07-06", "categories": []}
{"id": "2206.12569", "title": "Making Look-Ahead Active Learning Strategies Feasible with Neural\n  Tangent Kernels", "abstract": "  We propose a new method for approximating active learning acquisition\nstrategies that are based on retraining with hypothetically-labeled candidate\ndata points. Although this is usually infeasible with deep networks, we use the\nneural tangent kernel to approximate the result of retraining, and prove that\nthis approximation works asymptotically even in an active learning setup --\napproximating \"look-ahead\" selection criteria with far less computation\nrequired. This also enables us to conduct sequential active learning, i.e.\nupdating the model in a streaming regime, without needing to retrain the model\nwith SGD after adding each new data point. Moreover, our querying strategy,\nwhich better understands how the model's predictions will change by adding new\ndata points in comparison to the standard (\"myopic\") criteria, beats other\nlook-ahead strategies by large margins, and achieves equal or better\nperformance compared to state-of-the-art methods on several benchmark datasets\nin pool-based active learning.\n", "date": "2022-06-25", "categories": []}
{"id": "2207.01190", "title": "Pareto Optimization for Active Learning under Out-of-Distribution Data\n  Scenarios", "abstract": "  Pool-based Active Learning (AL) has achieved great success in minimizing\nlabeling cost by sequentially selecting informative unlabeled samples from a\nlarge unlabeled data pool and querying their labels from oracle/annotators.\nHowever, existing AL sampling strategies might not work well in\nout-of-distribution (OOD) data scenarios, where the unlabeled data pool\ncontains some data samples that do not belong to the classes of the target\ntask. Achieving good AL performance under OOD data scenarios is a challenging\ntask due to the natural conflict between AL sampling strategies and OOD sample\ndetection. AL selects data that are hard to be classified by the current basic\nclassifier (e.g., samples whose predicted class probabilities have high\nentropy), while OOD samples tend to have more uniform predicted class\nprobabilities (i.e., high entropy) than in-distribution (ID) data. In this\npaper, we propose a sampling scheme, Monte-Carlo Pareto Optimization for Active\nLearning (POAL), which selects optimal subsets of unlabeled samples with fixed\nbatch size from the unlabeled data pool. We cast the AL sampling task as a\nmulti-objective optimization problem, and thus we utilize Pareto optimization\nbased on two conflicting objectives: (1) the normal AL data sampling scheme\n(e.g., maximum entropy), and (2) the confidence of not being an OOD sample.\nExperimental results show its effectiveness on both classical Machine Learning\n(ML) and Deep Learning (DL) tasks.\n", "date": "2022-07-04", "categories": []}
{"id": "2206.12598", "title": "Mitigating sampling bias in risk-based active learning via an EM\n  algorithm", "abstract": "  Risk-based active learning is an approach to developing statistical\nclassifiers for online decision-support. In this approach, data-label querying\nis guided according to the expected value of perfect information for incipient\ndata points. For SHM applications, the value of information is evaluated with\nrespect to a maintenance decision process, and the data-label querying\ncorresponds to the inspection of a structure to determine its health state.\nSampling bias is a known issue within active-learning paradigms; this occurs\nwhen an active learning process over- or undersamples specific regions of a\nfeature-space, thereby resulting in a training set that is not representative\nof the underlying distribution. This bias ultimately degrades decision-making\nperformance, and as a consequence, results in unnecessary costs incurred. The\ncurrent paper outlines a risk-based approach to active learning that utilises a\nsemi-supervised Gaussian mixture model. The semi-supervised approach\ncounteracts sampling bias by incorporating pseudo-labels for unlabelled data\nvia an EM algorithm. The approach is demonstrated on a numerical example\nrepresentative of the decision processes found in SHM.\n", "date": "2022-06-25", "categories": []}
{"id": "2206.13391", "title": "Deep reinforced active learning for multi-class image classification", "abstract": "  High accuracy medical image classification can be limited by the costs of\nacquiring more data as well as the time and expertise needed to label existing\nimages. In this paper, we apply active learning to medical image\nclassification, a method which aims to maximise model performance on a minimal\nsubset from a larger pool of data. We present a new active learning framework,\nbased on deep reinforcement learning, to learn an active learning query\nstrategy to label images based on predictions from a convolutional neural\nnetwork. Our framework modifies the deep-Q network formulation, allowing us to\npick data based additionally on geometric arguments in the latent space of the\nclassifier, allowing for high accuracy multi-class classification in a\nbatch-based active learning setting, enabling the agent to label datapoints\nthat are both diverse and about which it is most uncertain. We apply our\nframework to two medical imaging datasets and compare with standard query\nstrategies as well as the most recent reinforcement learning based active\nlearning approach for image classification.\n", "date": "2022-06-20", "categories": []}
{"id": "2206.11616", "title": "Improving decision-making via risk-based active learning: Probabilistic\n  discriminative classifiers", "abstract": "  Gaining the ability to make informed decisions on operation and maintenance\nof structures provides motivation for the implementation of structural health\nmonitoring (SHM) systems. However, descriptive labels for measured data\ncorresponding to health-states of the monitored system are often unavailable.\nThis issue limits the applicability of fully-supervised machine learning\nparadigms for the development of statistical classifiers to be used in\ndecision-support in SHM systems. One approach to dealing with this problem is\nrisk-based active learning. In such an approach, data-label querying is guided\naccording to the expected value of perfect information for incipient data\npoints. For risk-based active learning in SHM, the value of information is\nevaluated with respect to a maintenance decision process, and the data-label\nquerying corresponds to the inspection of a structure to determine its health\nstate.\n  In the context of SHM, risk-based active learning has only been considered\nfor generative classifiers. The current paper demonstrates several advantages\nof using an alternative type of classifier -- discriminative models. Using the\nZ24 Bridge dataset as a case study, it is shown that discriminative classifiers\nhave benefits, in the context of SHM decision-support, including improved\nrobustness to sampling bias, and reduced expenditure on structural inspections.\n", "date": "2022-06-23", "categories": []}
{"id": "2206.08566", "title": "Active Data Discovery: Mining Unknown Data using Submodular Information\n  Measures", "abstract": "  Active Learning is a very common yet powerful framework for iteratively and\nadaptively sampling subsets of the unlabeled sets with a human in the loop with\nthe goal of achieving labeling efficiency. Most real world datasets have\nimbalance either in classes and slices, and correspondingly, parts of the\ndataset are rare. As a result, there has been a lot of work in designing active\nlearning approaches for mining these rare data instances. Most approaches\nassume access to a seed set of instances which contain these rare data\ninstances. However, in the event of more extreme rareness, it is reasonable to\nassume that these rare data instances (either classes or slices) may not even\nbe present in the seed labeled set, and a critical need for the active learning\nparadigm is to efficiently discover these rare data instances. In this work, we\nprovide an active data discovery framework which can mine unknown data slices\nand classes efficiently using the submodular conditional gain and submodular\nconditional mutual information functions. We provide a general algorithmic\nframework which works in a number of scenarios including image classification\nand object detection and works with both rare classes and rare slices present\nin the unlabeled set. We show significant accuracy and labeling efficiency\ngains with our approach compared to existing state-of-the-art active learning\napproaches for actively discovering these rare classes and slices.\n", "date": "2022-06-17", "categories": []}
{"id": "2206.08061", "title": "Active Nearest Neighbor Regression Through Delaunay Refinement", "abstract": "  We introduce an algorithm for active function approximation based on nearest\nneighbor regression. Our Active Nearest Neighbor Regressor (ANNR) relies on the\nVoronoi-Delaunay framework from computational geometry to subdivide the space\ninto cells with constant estimated function value and select novel query points\nin a way that takes the geometry of the function graph into account. We\nconsider the recent state-of-the-art active function approximator called DEFER,\nwhich is based on incremental rectangular partitioning of the space, as the\nmain baseline. The ANNR addresses a number of limitations that arise from the\nspace subdivision strategy used in DEFER. We provide a computationally\nefficient implementation of our method, as well as theoretical halting\nguarantees. Empirical results show that ANNR outperforms the baseline for both\nclosed-form functions and real-world examples, such as gravitational wave\nparameter inference and exploration of the latent space of a generative model.\n", "date": "2022-06-16", "categories": []}
{"id": "2206.06276", "title": "On the reusability of samples in active learning", "abstract": "  An interesting but not extensively studied question in active learning is\nthat of sample reusability: to what extent can samples selected for one learner\nbe reused by another? This paper explains why sample reusability is of\npractical interest, why reusability can be a problem, how reusability could be\nimproved by importance-weighted active learning, and which obstacles to\nuniversal reusability remain. With theoretical arguments and practical\ndemonstrations, this paper argues that universal reusability is impossible.\nBecause every active learning strategy must undersample some areas of the\nsample space, learners that depend on the samples in those areas will learn\nmore from a random sample selection. This paper describes several experiments\nwith importance-weighted active learning that show the impact of the\nreusability problem in practice. The experiments confirmed that universal\nreusability does not exist, although in some cases -- on some datasets and with\nsome pairs of classifiers -- there is sample reusability. Finally, this paper\nexplores the conditions that could guarantee the reusability between two\nclassifiers.\n", "date": "2022-06-13", "categories": []}
{"id": "2206.05009", "title": "Weighted Ensembles for Active Learning with Adaptivity", "abstract": "  Labeled data can be expensive to acquire in several application domains,\nincluding medical imaging, robotics, and computer vision. To efficiently train\nmachine learning models under such high labeling costs, active learning (AL)\njudiciously selects the most informative data instances to label on-the-fly.\nThis active sampling process can benefit from a statistical function model,\nthat is typically captured by a Gaussian process (GP). While most GP-based AL\napproaches rely on a single kernel function, the present contribution advocates\nan ensemble of GP models with weights adapted to the labeled data collected\nincrementally. Building on this novel EGP model, a suite of acquisition\nfunctions emerges based on the uncertainty and disagreement rules. An\nadaptively weighted ensemble of EGP-based acquisition functions is also\nintroduced to further robustify performance. Extensive tests on synthetic and\nreal datasets showcase the merits of the proposed EGP-based approaches with\nrespect to the single GP-based AL alternatives.\n", "date": "2022-06-10", "categories": []}
{"id": "2206.03288", "title": "Collaborative Intelligence Orchestration: Inconsistency-Based Fusion of\n  Semi-Supervised Learning and Active Learning", "abstract": "  While annotating decent amounts of data to satisfy sophisticated learning\nmodels can be cost-prohibitive for many real-world applications. Active\nlearning (AL) and semi-supervised learning (SSL) are two effective, but often\nisolated, means to alleviate the data-hungry problem. Some recent studies\nexplored the potential of combining AL and SSL to better probe the unlabeled\ndata. However, almost all these contemporary SSL-AL works use a simple\ncombination strategy, ignoring SSL and AL's inherent relation. Further, other\nmethods suffer from high computational costs when dealing with large-scale,\nhigh-dimensional datasets. Motivated by the industry practice of labeling data,\nwe propose an innovative Inconsistency-based virtual aDvErsarial Active\nLearning (IDEAL) algorithm to further investigate SSL-AL's potential\nsuperiority and achieve mutual enhancement of AL and SSL, i.e., SSL propagates\nlabel information to unlabeled samples and provides smoothed embeddings for AL,\nwhile AL excludes samples with inconsistent predictions and considerable\nuncertainty for SSL. We estimate unlabeled samples' inconsistency by\naugmentation strategies of different granularities, including fine-grained\ncontinuous perturbation exploration and coarse-grained data transformations.\nExtensive experiments, in both text and image domains, validate the\neffectiveness of the proposed algorithm, comparing it against state-of-the-art\nbaselines. Two real-world case studies visualize the practical industrial value\nof applying and deploying the proposed data sampling algorithm.\n", "date": "2022-06-07", "categories": []}
{"id": "2606.22002", "title": "One-Shot Data Selection for Medical Image Classification via Graph Coverage", "abstract": "Training medical image classifiers on entire datasets is wasteful when annotation budgets are limited: not all samples contribute equally, yet acquiring expert labels is expensive. Active learning reduces annotation cost through iterative querying, but assumes repeated access to an oracle and requires multiple rounds of model training. One-shot geometry-based methods such as facility location avoid retraining but operate on pairwise distances that ignore the local structure of the data manifold. We propose a graph-based one-shot selection method that operates entirely on frozen foundation model embeddings. Given embeddings from a pretrained encoder, we construct a k-nearest neighbor graph over all training samples and derive a two-term coverage kernel from the heat diffusion kernel, capturing both direct and two-hop neighborhood relationships. Greedy facility location on this kernel selects class-balanced subsets that maximize coverage of the data manifold. The two-term kernel matches the full spectral heat kernel in selection behavior while reducing computation to sparse matrix operations with a single hyperparameter. We evaluate on five MedMNIST datasets spanning histopathology, radiology, and microscopy, comparing against both training-dynamics and geometry-based baselines. Our method achieves the highest balanced accuracy on nine of ten dataset-ratio conditions, with the largest gains on class-imbalanced datasets where global graph construction captures cross-class structure that per-class methods miss, all without any model training during selection. Code is available at https://github.com/zahiriddin-rustamov/graph-coverage-selection.", "date": "2026-06-20", "categories": []}
{"id": "2606.28659", "title": "Transformer-Based Active Learning for Data-Efficient Vaccine Epitope Selection in PRRS", "abstract": "High-fidelity molecular docking simulations can produce biologically relevant estimates of epitope-receptor binding affinity but are computationally expensive and therefore limit the number of candidates that can be screened for vaccine design. In this work, we evaluate machine learning (ML) approaches where variants of active learning are used to classify instances of high binding affinity between 9-mer epitopes and a well-conserved swine leukocyte antigen (SLA) receptor in the context of Porcine Reproductive and Respiratory Syndrome (PRRS). We use an internally generated dataset of 80 epitope-SLA docking affinities, each requiring more than 48 hours of high-performance computing (HPC). Multiple model families (linear, MLP, CNN, and a small transformer) are trained under strict low-data conditions within a pool-based active learning loop. In each case, optimal model configurations are identified by conducting large-scale hyperparameter optimization over the combined space of model architecture, training configuration, acquisition policy, and ensemble decision rules. To mitigate the effects of data subsample selection, each candidate configuration is evaluated by averaging performance over many randomized and balanced training and validation data subsets. Across experiments, transformer-based sequence models consistently emerged as the best-performing architecture, with active incremental learning yielding significant improvement over a baseline random sample acquisition strategy. Under moderate training data availability (N=30), the optimized ML-model configuration outperforms a standard baseline trained on twice the amount of data. Under higher training data availability (N=60), the same configuration achieves a peak accuracy of 86.8%, consistent with an upper bound of 85% classification accuracy based on two independent estimates of conformational noise.", "date": "2026-06-27", "categories": []}
{"id": "2607.00144", "title": "A Mechanism-Driven Theory of Phase Transitions in Active Learning", "abstract": "Active learning (AL) performance is known to be budget-dependent, yet regimes are typically defined by heuristic label counts that fail to generalize across datasets or architectures. We characterize AL dynamics by reframing budget regimes as shifts in the dominant generalization mechanism. By reinterpreting PAC-style risk components as dynamic interacting terms, we prove that dominance shifts are structurally unavoidable, creating a moving bottleneck for generalization. We operationalize this using measurable proxies and a segmented regression procedure to identify a tripartite taxonomy: data-driven, transition, and model-driven phases. Our framework explains the long-standing observation that representativeness, coverage, and uncertainty strategies excel at different stages. Experiments across natural and medical imaging show that AL efficiency depends on the alignment between the strategy's inductive bias and the active bottleneck. Moreover, self-supervised representation shift transitions earlier along the labeling trajectory, highlighting the role of representation quality in shaping AL dynamics. Overall, this work provides a unified framework for the next generation of transition-aware AL algorithms.", "date": "2026-06-30", "categories": []}