text string | arxiv_id string | url string | timestamp string | yymm string | language string | Qwen3_token_count int64 | uuid string | start_index int64 | input_ids list |
|---|---|---|---|---|---|---|---|---|---|
idea in \cite{athalye2018synthesizing}.
\subsection{Defending Against Adversarial Examples}
Conversely, many methods have been proposed recently to defend against adversarial examples. \cite{goodfellow2014explaining,kurakin2016scale} proposed to inject adversarial examples into the training data to increase the netw... | 1803.06978 | https://arxiv.org/abs/1803.06978 | 2019-06-04 02:11:33 | 1803 | en | 4,096 | 7a303b5b-182f-44fc-ab75-7fdb677ed650 | 1,883 | [
4522,
304,
1124,
67075,
90,
587,
5774,
68,
17,
15,
16,
23,
20339,
6375,
4849,
7810,
4710,
59,
68874,
90,
2620,
2459,
29141,
2410,
3004,
42721,
25311,
532,
1109,
3004,
974,
11,
1657,
5413,
614,
1012,
10981,
5926,
311,
10505,
2348,
2841... |
to 1. An instructor understands that as a first step, a student might use a \texttt{for} loop or manually write ten \texttt{print} statements. Given that the student uses a loop, they could increment up or down. And given that they increment down, their loop comparison could be correct or off-by-one. At each of these... | 1905.09916 | https://arxiv.org/abs/1905.09916 | 2021-03-25 01:00:52 | 1905 | en | 4,096 | 85ccd202-cdd3-459f-8324-6b2456e8fec1 | 3,185 | [
311,
220,
16,
13,
1527,
32215,
30769,
429,
438,
264,
1156,
3019,
11,
264,
5458,
2578,
990,
264,
1124,
1318,
5566,
90,
1958,
92,
6337,
476,
20083,
3270,
5779,
1124,
1318,
5566,
90,
1350,
92,
12239,
13,
16246,
429,
279,
5458,
5711,
26... |
that we minimize the error metric defined above, $err(\hat{y}_i, y_i)$. For supervised approaches, we split $D$ into a training ($D_\text{train}$) and test set ($D_\text{test}$) via a 80-20 ratio. For unsupervised methods, we evaluate on the entire set $D$.
\begin{lstlisting}[breaklines=true,caption={Example $x_i$ fr... | 1809.01357 | https://arxiv.org/abs/1809.01357 | 2018-12-18 02:21:51 | 1809 | en | 4,096 | 04d851c5-ed96-4071-a89c-c279227410af | 3,178 | [
429,
582,
29337,
279,
1465,
18266,
4512,
3403,
11,
400,
615,
11520,
9198,
90,
88,
19444,
72,
11,
379,
5318,
8,
12947,
1752,
58989,
19827,
11,
582,
6718,
400,
35,
3,
1119,
264,
4862,
1711,
35,
56014,
1318,
90,
10397,
31716,
8,
323,
... |
T \cdot D_e^\star$
\STATE Choose $P_k = D_k D^{\dagger}$ , i.e. minimizing $||D_k - P_k \cdot D||_F$
\end{algorithmic}
\caption{The RCNCM-Duarte algorithm. The algorithm finds an optimized acquisition matrix from the solution of a rank-constrained nearest correlation matrix problem.}
\label{algo_RCNCM_Du}
\end{figure}
... | 1309.3676 | https://arxiv.org/abs/1309.3676 | 2013-09-17 02:04:21 | 1309 | en | 4,096 | e7e35622-9c97-4f57-b543-b66679634bd1 | 8,654 | [
51,
1124,
50853,
422,
2204,
24884,
11870,
25046,
59,
24628,
22201,
400,
47,
4698,
284,
422,
4698,
422,
61,
35702,
67,
10114,
31716,
1154,
600,
1734,
13,
76291,
400,
8484,
35,
4698,
481,
393,
4698,
1124,
50853,
422,
8484,
62,
37,
25046... |
detection performance (please, see the source code).}}
We emphasize that removing the entropic scale after the training does not affect the solution's ability to represent the prior knowledge available, as it does not change the predictions. Therefore, the expression for the probabilities with the entropic scale remo... | 2006.04005 | https://arxiv.org/abs/2006.04005 | 2021-08-06 02:01:22 | 2006 | en | 4,096 | 1a8f179d-8667-4191-bda7-a7e49ce0d3fb | 8,810 | [
17984,
5068,
320,
30021,
11,
1490,
279,
2530,
2038,
568,
47449,
1654,
45932,
429,
17592,
279,
1197,
43936,
5452,
1283,
279,
4862,
1558,
537,
7802,
279,
6291,
594,
5726,
311,
4009,
279,
4867,
6540,
2500,
11,
438,
432,
1558,
537,
2297,
... |
2} & \textbf{1.9E-03} & \textbf{2.3E-03} \\ \midrule
\multirow{6}{*}{AUC-ROC} & BR & \textbf{9.4E-04} & \textbf{8.3E-04} & \textbf{9.4E-04} & \textbf{4.0E-03} & \textbf{4.0E-03} & \textbf{7.4E-04} \\
& MLkNN & \textbf{7.4E-04} & \textbf{9.4E-04} & \textbf{1.7E-03} & \textbf{4.0E-03} & \textbf{4.0E-03} & \textbf{7.4E-... | 2005.03240 | https://arxiv.org/abs/2005.03240 | 2020-05-20 02:13:58 | 2005 | en | 4,096 | 496396a5-a46b-4ea5-83d9-e280a01207cf | 19,871 | [
17,
92,
609,
1124,
1318,
13233,
90,
16,
13,
24,
36,
12,
15,
18,
92,
609,
1124,
1318,
13233,
90,
17,
13,
18,
36,
12,
15,
18,
92,
24984,
220,
1124,
16106,
12937,
198,
59,
26268,
651,
90,
21,
15170,
9,
15170,
32,
5459,
12,
73645,... |
approximation in the temporal domain:
\[
\Sigma(\tau) \approx \int \exp(-i\tau \omega) M \frac{1}{n} \sum_{k}^{\ell} \gamma K(\gamma (\omega - \mu_k)) d\omega
= C_\mathrm{SM}(\tau; K,M,\mu,\gamma).
\]
Since the approximation error in the spectral domain was small, the approximation error in the k... | 2003.05554 | https://arxiv.org/abs/2003.05554 | 2021-02-08 02:01:23 | 2003 | en | 4,096 | 0f0073c5-badd-44ec-b9b2-41271b3bb39e | 5,488 | [
56204,
304,
279,
35915,
7947,
510,
262,
1124,
9640,
262,
1124,
62016,
11520,
29143,
8,
1124,
48053,
1124,
396,
1124,
4580,
4080,
72,
59,
29143,
1124,
32696,
8,
386,
1124,
37018,
90,
16,
15170,
77,
92,
1124,
1242,
15159,
74,
92,
61,
... |
5batch} but with different flavors. Both approaches can be considered to tackle the issue of internal covariate shift for accelerating the convergence. Batch normalization tackles the issue by normalizing the output of neurons to zero mean and unit variance and then performing dropout independently~\footnote{The autho... | 1602.02220 | https://arxiv.org/abs/1602.02220 | 2016-12-06 02:06:13 | 1602 | en | 4,096 | cb248f6b-786e-465e-8d09-d5e0498bc393 | 572 | [
20,
14049,
92,
714,
448,
2155,
31423,
13,
11733,
19827,
646,
387,
6509,
311,
21403,
279,
4265,
315,
5306,
21836,
49659,
6407,
369,
68641,
220,
279,
55441,
13,
33904,
48723,
42481,
279,
4265,
553,
4622,
4849,
279,
2550,
315,
33213,
311,
... |
olumn{3}{c}{\textbf{Dataset}}\\
\cmidrule{2-4}
\textbf{Statistics} & \textbf{HS} & {\textbf{ATIS}} & {\textbf{JOBS}}\\
\midrule
\# Train & 533 & 4,434 & 500\\
\# Dev & 66 & 491 & -\\
\# Test & 66 & 448 & 140\\
\midrule
Avg. tokens in description & 35.0 & 10.6 & 8.7\\
Max. tokens in d... | 1811.06837 | https://arxiv.org/abs/1811.06837 | 2018-11-19 02:11:32 | 1811 | en | 4,096 | b41ccd08-1cdb-4bd8-a3e9-2dde454e311c | 6,600 | [
1286,
90,
18,
15170,
66,
15170,
59,
1318,
13233,
90,
33363,
3417,
3422,
198,
298,
197,
59,
6226,
307,
12937,
90,
17,
12,
19,
532,
298,
197,
59,
1318,
13233,
90,
38599,
92,
609,
1124,
1318,
13233,
90,
11961,
92,
220,
609,
28152,
13... |
on a human-collected dataset of paired demonstrations, which consists of pairs of optimal trajectories corresponding to the same task (following the standard meta-imitation learning set-up in prior work~\cite{duan2017one,finn2017one}).
We then use this policy to collect trials autonomously. For each such trial, a user... | 2003.02636 | https://arxiv.org/abs/2003.02636 | 2020-03-06 02:14:33 | 2003 | en | 4,096 | 25462104-5245-45b6-b3d4-9e4efd9cde58 | 1,634 | [
389,
264,
3738,
19459,
2209,
10337,
315,
34426,
43795,
11,
892,
17167,
315,
13530,
315,
22823,
85548,
12159,
311,
279,
1852,
3383,
320,
42918,
279,
5297,
8823,
12,
2353,
367,
6832,
738,
5239,
304,
4867,
975,
93,
59,
67075,
90,
1054,
2... |
_{\rm{f-en}}(\mathbf{H}, \mathbf{\Theta}_{en})+\Delta \mathbf{M}, \mathbf{\Theta}_{p}) \nonumber \\
& + \lambda \text{MSE} \bigg(f_{\rm{f-de}}\big(f_{\rm{f-en}}(\mathbf{H}, \mathbf{\Theta}_{en})+\Delta \mathbf{M}, \mathbf{\Theta}_{de}\big), \mathbf{H}\bigg) \Bigg),
\end{align}
where
\begin{equation}\label{Loss1}
... | 2003.04684 | https://arxiv.org/abs/2003.04684 | 2020-09-09 02:14:28 | 2003 | en | 4,096 | d5a352a2-5d74-46fe-9ed5-d9f0287907a1 | 5,772 | [
15159,
59,
8719,
90,
69,
20767,
3417,
11520,
10374,
13233,
90,
39,
2137,
1124,
10374,
13233,
35702,
44314,
51535,
268,
5410,
41715,
20277,
1124,
10374,
13233,
90,
44,
2137,
1124,
10374,
13233,
35702,
44314,
51535,
79,
5410,
1124,
6280,
89... |
consistent_local}) result from a failure to average feature impacts over all orderings. 2) For an infinite ensemble of random fully-developed trees on binary features, Saabas values effectively consider all orderings and so converge to the SHAP values. In practice, however, tree ensemble models are not infinite, random... | 1905.04610 | https://arxiv.org/abs/1905.04610 | 2019-05-14 02:16:48 | 1905 | en | 4,096 | d87a19d3-fdb7-4780-ac2c-feef35649346 | 3,994 | [
78399,
13564,
5410,
1102,
504,
264,
7901,
311,
5461,
4565,
24905,
916,
678,
1973,
819,
13,
220,
17,
8,
1752,
458,
23809,
39026,
315,
4194,
7225,
47961,
291,
12408,
389,
7868,
4419,
11,
15854,
370,
300,
2750,
13444,
2908,
678,
1973,
81... |
the underlying game and converges to stationary NE. To the best of our knowledge, this is the first algorithm to do so, with proven convergence.
\end{remark}
\section{Formal Definitions}
\label{sect:prelims}
A stochastic game can be seen to be an extension of the single-agent Markov decision process. A discounted r... | 1401.2086 | https://arxiv.org/abs/1401.2086 | 2015-07-06 02:00:43 | 1401 | en | 4,096 | b257a315-f7b3-473f-908f-45e43d003def | 5,214 | [
279,
16533,
1809,
323,
19313,
4188,
311,
52635,
7856,
13,
2014,
279,
1850,
315,
1039,
6540,
11,
419,
374,
279,
1156,
12111,
311,
653,
773,
11,
448,
16622,
55441,
624,
59,
408,
90,
37448,
13246,
59,
2809,
90,
1838,
278,
46513,
532,
5... |
the number $K$ of distinct functions $f_k(\cdot)$ is unknown must also be inferred from the data.
This is achieved by formulating an alignment objective that pushes the uncorrupted sequences $\mathbf{f}_j$ into $K$ groups where each group corresponds to one latent function $f_k(\cdot)$, and the sequences within each ... | 1905.12930 | https://arxiv.org/abs/1905.12930 | 2020-02-26 02:11:18 | 1905 | en | 4,096 | ae81d2b5-c32d-41c8-83da-81e22cd41e51 | 4,115 | [
279,
1372,
400,
42,
3,
315,
12460,
5746,
400,
69,
4698,
11520,
50853,
15087,
374,
9788,
1969,
1083,
387,
67595,
504,
279,
821,
13,
715,
1986,
374,
16994,
553,
1352,
15486,
458,
17189,
16538,
429,
40702,
279,
650,
6005,
85954,
23700,
5... |
}\}$ and $\boldsymbol L$ is
\begin{align}
p(&\{\gamma_{R}\}_{R \in \mathcal A}, \boldsymbol L |\boldsymbol l, \boldsymbol w) \propto \\
&\left(\textstyle\prod_{R \in \mathcal A} \gamma_{R}^{\mu_{R}(\Theta) + l_{l_{\cdot R}} - 1}e^{-\gamma_{R}}\right)\left(\textstyle\prod_{j} L_{j}^{l_{j\cdot}-1}e^{-\sum_{R \in \mat... | 1407.0044 | https://arxiv.org/abs/1407.0044 | 2014-07-02 02:01:01 | 1407 | en | 4,096 | 36a48a7e-4ef8-42bf-98dd-73503154655c | 16,897 | [
11035,
31716,
323,
57960,
14824,
18785,
444,
3,
374,
715,
59,
7265,
90,
6612,
92,
2303,
79,
2099,
59,
35702,
32214,
15159,
49,
11035,
51535,
49,
1124,
258,
1124,
10374,
5416,
362,
2137,
1124,
14824,
18785,
444,
63596,
14824,
18785,
326,... |
function.
\subsection{Notations}
Throughout the paper, vectors and matrices are denoted by boldface, e.g. ${\boldsymbol x},{\boldsymbol y}, {\boldsymbol z},\dots$.
We reserve the upper case, e.g. ${\boldsymbol A},{\boldsymbol B},\dots$, for $n\times n$ matrices.
Their coordinates are denoted by non-bold font, e.g.... | 1603.04064 | https://arxiv.org/abs/1603.04064 | 2016-03-15 01:10:59 | 1603 | en | 4,096 | e3a36bc3-3134-4b8b-9068-559a58620e87 | 3,075 | [
729,
382,
59,
68874,
90,
2623,
804,
630,
59005,
279,
5567,
11,
22879,
220,
323,
35195,
525,
3371,
9253,
553,
13939,
1564,
11,
384,
1302,
13,
3570,
59,
14824,
18785,
856,
16561,
59,
14824,
18785,
379,
2137,
28152,
14824,
18785,
1147,
2... |
0.015 \\ \midrule
\textbf{OKGAN}
& \textbf{25.0} & 86.2 & \textbf{0.006}
& \textbf{8.0} & 95.3 & \textbf{0.002}
& \textbf{1.0} & 98.1 & \textbf{0.0003} \\
\bottomrule
\end{tabular}
\end{table}
As you can see from Table \ref{table1} and Appx. \ref{appendix_2d_synthetic}, OKGAN shows the best performance overall in ... | 2006.11432 | https://arxiv.org/abs/2006.11432 | 2020-06-23 02:05:04 | 2006 | en | 4,096 | 6811be55-236c-4341-9562-099c3e386f22 | 6,061 | [
220,
15,
13,
15,
16,
20,
24984,
1124,
16106,
12937,
198,
59,
1318,
13233,
90,
3925,
58487,
92,
715,
5,
1124,
1318,
13233,
90,
17,
20,
13,
15,
92,
609,
220,
23,
21,
13,
17,
609,
1124,
1318,
13233,
90,
15,
13,
15,
15,
21,
532,
... |
\mathbf \Phi}}({\bf x}|{\bf y}) \right]\\
&= \sum_i \sum_{x_i} q_{{\mathbf \Phi}}(x_{i}|{\bf y}) \log p_i(x_i) + {\cal H} \left[ q_{{\mathbf \Phi}}({\bf x}|{\bf y}) \right]\\
&= \sum_i \sum_{x \in \left\{-1,1\right\}} q_{{\mathbf \Phi},i}(x|{\bf y}) \log p_i(x) + {\cal H} \left[ q_{{\mathbf \Phi}}({\bf x}|{\bf y}) \rig... | 1905.08795 | https://arxiv.org/abs/1905.08795 | 2020-04-14 02:20:59 | 1905 | en | 4,096 | ce1adce0-51fc-4e7a-bbce-90951c503948 | 18,428 | [
59,
10374,
13233,
1124,
45167,
3417,
2306,
59,
13233,
856,
52398,
35702,
13233,
379,
5410,
1124,
1291,
86920,
198,
5,
28,
1124,
1242,
5318,
1124,
1242,
15159,
87,
5318,
92,
2804,
62,
2979,
59,
10374,
13233,
1124,
45167,
3417,
7,
87,
1... |
able to extract large quantities of labelled information for many problems of interest, there remains a significant proportion of tasks for which ``big data'' simply isn't available at this time, which makes it extremely difficult to fully exploit a deep CNN architecture and properly learn generalisable features of th... | 1610.00163 | https://arxiv.org/abs/1610.00163 | 2016-10-18 02:12:41 | 1610 | en | 4,096 | b7447b77-b1b5-4582-95d2-415e77174707 | 168 | [
2952,
311,
8649,
3460,
32676,
315,
60236,
1995,
369,
1657,
5322,
315,
2734,
11,
1052,
8458,
264,
5089,
21117,
315,
9079,
369,
892,
9902,
16154,
821,
4605,
4936,
4436,
944,
2500,
518,
419,
882,
11,
892,
3643,
432,
9016,
5000,
311,
7225... |
{equation}
Here $\lambda$ is a trade-off constant and $\lambda$ $> 0$. This last term is introduced to make sure that the synthetic image produced by the generator is not too much deviated from the real image. It can be considered as simple L1 loss function, denoted as:
\begin{equation}
\label{eq:a2}
L_{DEV} \; (G_{\th... | 1810.00551 | https://arxiv.org/abs/1810.00551 | 2018-10-02 02:18:53 | 1810 | en | 4,096 | 288707e4-3b3e-4011-800d-6e7a66d8d321 | 2,135 | [
90,
25310,
367,
532,
8420,
57960,
12935,
3,
374,
264,
6559,
12462,
6783,
323,
57960,
12935,
3,
400,
29,
220,
15,
12947,
1096,
1537,
4647,
374,
11523,
311,
1281,
2704,
429,
279,
27268,
2168,
8947,
553,
279,
13823,
374,
537,
2238,
1753,... |
{Experiments}
\label{sec:experiments}
\textbf{Compared Methods.}
We make benchmarks of our proposed method aggregation of multiple knockoffs
(AKO) with \( B=25, \gamma=0.3 \) and vanilla knockoff (KO), along with other
recent methods for controlling FDR in high-dimensional settings, mentioned in
Section \ref{ssec:relat... | 2002.09269 | https://arxiv.org/abs/2002.09269 | 2020-06-26 02:16:01 | 2002 | en | 4,096 | 2844f666-3615-44a7-833d-4ab7c1267bef | 10,381 | [
90,
840,
49632,
532,
59,
1502,
90,
5024,
25,
327,
49632,
532,
59,
1318,
13233,
90,
1092,
7212,
18801,
13,
532,
1654,
1281,
62019,
315,
1039,
10981,
1714,
51629,
315,
5248,
14127,
32483,
198,
4346,
54947,
8,
448,
17767,
425,
28,
17,
... |
mathbb{Z}}),\tilde{{\mathbb{Z}}})$ among
$$\{\widehat{R}(a({\mathbb{Z}}^{\pi_1}),\tilde{{\mathbb{Z}}}^{\pi_1}),\ldots,\widehat{R}(a({\mathbb{Z}}^{\pi_B}),\tilde{{\mathbb{Z}}}^{\pi_B})\},$$
where $a \in {\mathcal{A}}$ is any prediction method.
We then reject the
null hypothesis if this statistic is small.
The intui... | 1908.00105 | https://arxiv.org/abs/1908.00105 | 2019-08-02 02:04:06 | 1908 | en | 4,096 | d1aafe78-36eb-4821-a375-14be01c5df2e | 2,640 | [
10374,
6066,
90,
57,
3417,
701,
59,
1646,
450,
2979,
59,
10374,
6066,
90,
57,
3417,
5410,
3,
4221,
198,
262,
26107,
59,
35702,
9150,
9198,
90,
49,
25547,
64,
2306,
59,
10374,
6066,
90,
57,
3417,
61,
35702,
2493,
62,
16,
38842,
59,... |
shape priors learned using statistical shape models, to jointly and simultaneously predict the parameters of 3D cardiac shapes, instead of a pixel-wise classification across each 2D slice. Another significant aspect of this work is the integration of patient metadata into the process of shape prediction using a Multil... | 1907.01913 | https://arxiv.org/abs/1907.01913 | 2019-07-04 02:14:04 | 1907 | en | 4,096 | 75f0a0d5-6f2e-46d9-ba57-6f32f3f6e3e6 | 513 | [
6083,
12493,
1087,
9498,
1667,
28464,
6083,
4119,
11,
311,
52158,
323,
24303,
7023,
279,
5029,
315,
220,
18,
35,
46245,
20816,
11,
4518,
315,
264,
12955,
44439,
23850,
3941,
1817,
220,
17,
35,
15983,
13,
13293,
5089,
12893,
315,
419,
... |
:eval}) have millions of binary features, rather than few categorical ones with many values.
\item \cite{naivesub} extend their method to multiple categorical variables with a heuristic, but it does not perform well when significantly reducing dimension (Sec.~\ref{sec:global-submod}).
\end{enumerate}
\section{Chro... | 2006.03779 | https://arxiv.org/abs/2006.03779 | 2020-06-09 02:06:20 | 2006 | en | 4,096 | 98002130-c76b-459b-a31d-93d9c23bba9e | 2,671 | [
25,
14170,
5410,
614,
11728,
315,
7868,
4419,
11,
4751,
1091,
2421,
69536,
6174,
448,
1657,
2750,
624,
262,
1124,
1203,
1124,
67075,
90,
3376,
1886,
392,
92,
13036,
862,
1714,
311,
5248,
69536,
7332,
448,
264,
66609,
11,
714,
432,
155... |
end{document}
\@startsection{section}{1}{\z@}{-0.12in}{0.02in{\@startsection{section}{1}{\z@}{-0.12in}{0.02in}
{\@setsize\large{14pt}\xiipt\@xiipt\bf\raggedright}}
\def\@startsection{subsection}{2}{\z@}{-0.10in}{0.01in{\@startsection{subsection}{2}{\z@}{-0.10in}{0.01in}
{\normalsize\bf\rag... | 1802.07182 | https://arxiv.org/abs/1802.07182 | 2018-02-22 02:07:05 | 1802 | en | 4,096 | d4f6f970-74de-4153-97f3-9f7880284149 | 18,628 | [
408,
90,
6062,
630,
59,
31,
2468,
2809,
90,
2809,
15170,
16,
15170,
59,
89,
31,
15170,
12,
15,
13,
16,
17,
258,
15170,
15,
13,
15,
17,
258,
35702,
31,
2468,
2809,
90,
2809,
15170,
16,
15170,
59,
89,
31,
15170,
12,
15,
13,
16,
... |
/qualitative/kodim17/proposed.pdf}
\caption{Proposed (SGA); \\ 0.13~BPP ($\text{PSNR}\!=\!31.2$)} %
\end{subfigure}\hfill
\begin{subfigure}[t]{.24\textwidth}
\centering
\includegraphics[width=\linewidth]{figs/qualitative/kodim17/baseline.pdf}
\caption{\citep{minnen2018joint}; \\ 0.12 BPP ($\text{PSNR}\!=\!29.8... | 2006.04240 | https://arxiv.org/abs/2006.04240 | 2021-01-11 02:09:37 | 2006 | en | 4,096 | 2a80eb2e-4b7e-4338-bddf-7ee1f47865ca | 13,722 | [
14,
1751,
21312,
14109,
347,
318,
16,
22,
18008,
3865,
15995,
532,
220,
1124,
23347,
90,
2008,
19586,
320,
7783,
32,
1215,
24984,
220,
15,
13,
16,
18,
93,
33,
4406,
220,
1711,
59,
1318,
90,
5012,
26319,
11035,
5824,
59,
0,
18,
16,... |
_3$), which are intermediate states.}
\vspace{-15pt}
\label{System_model}
\end{figure*}
\subsection{Graph-Constrained Changepoint Detection Model}
Let us assume a constrained changepoint model represented with a graph $G=(V,E)$, where the vertices $V\in\{1,\dots,|V|\}$ represent the hidden states or ... | 2004.13558 | https://arxiv.org/abs/2004.13558 | 2020-04-29 02:18:04 | 2004 | en | 4,096 | f321650f-2b17-449e-90b5-abb9d5c3689c | 1,709 | [
62,
18,
3,
701,
892,
525,
28439,
5302,
13,
532,
414,
1124,
85,
8746,
19999,
16,
20,
417,
532,
414,
1124,
1502,
90,
2320,
5047,
532,
256,
1124,
408,
90,
17781,
9,
3733,
59,
68874,
90,
11212,
57748,
57727,
43372,
747,
781,
57353,
49... |
2\nonumber\\
&\stackrel{(c)}{\leq}& \|\mathbf{x}_1-\mathbf{x}_2\|_2+\prod_{i=1}^{d}\|\mathbf{J}_i\|_{2,2}\|\mathbf{y}_1-\mathbf{y}_2\|_2\nonumber\\
&\stackrel{(d)}{\leq}& \left(1+\prod_{i=1}^{d}\|\mathbf{J}_i\|_{2,2}\right)\rho(\mathbf{s}_1,\mathbf{s}_2)
\end{IEEEeqnarray}
The inequalities $(a),(b)$ and $(c)$ hold ... | 1901.11352 | https://arxiv.org/abs/1901.11352 | 2019-02-01 02:13:20 | 1901 | en | 4,096 | df1727c1-b7b6-46c5-bb2e-c4462a8065a8 | 10,278 | [
17,
1699,
263,
890,
3422,
198,
197,
5,
59,
7693,
3748,
96065,
66,
9139,
35702,
273,
80,
24778,
1124,
22957,
10374,
13233,
45340,
19444,
16,
30529,
10374,
13233,
45340,
19444,
17,
59,
91,
62,
17,
41715,
19748,
15159,
72,
28,
16,
92,
... |
} & \num{0.79} & \num{0.04} \\
& AstroPh & \num{0.840} & \num{0.84} & \num{0.30} \\
\midrule
\multirow{3}{*}{\texttt{INS}\xspace}
& Amazon & \num{0.796} & \num{0.80} & \num{0.18} \\
& CondMat & \num{0.960} & \num{0.96} & \num{0.04} \\
& AstroPh & \num{0.965} & \num{0.97} & \num{0.02} \\
\bottomrule
\end{tabula... | 1902.06679 | https://arxiv.org/abs/1902.06679 | 2020-06-08 02:00:34 | 1902 | en | 4,096 | 41ad0010-9180-4f6a-805d-b9299806a6f2 | 4,837 | [
92,
609,
1124,
2413,
90,
15,
13,
22,
24,
92,
609,
1124,
2413,
90,
15,
13,
15,
19,
92,
90155,
197,
5,
64129,
3357,
197,
197,
5,
1124,
2413,
90,
15,
13,
23,
19,
15,
92,
609,
1124,
2413,
90,
15,
13,
23,
19,
92,
609,
1124,
241... |
-step
visual captioning process, independent of the parameterization in Section \ref{sec_parameterization}.
In practice, however, one needs to take into account several contributing factors to
the oracle.
Firstly, atoms, or visual concepts, may be defined as 1-gram words, 2-gram phrases and so on.
Arguably a mixture... | 1511.04590 | https://arxiv.org/abs/1511.04590 | 2016-09-15 02:07:12 | 1511 | en | 4,096 | b842da1f-d002-40b5-8c0b-7032c88d14ff | 3,850 | [
29208,
198,
29218,
17256,
287,
1882,
11,
9489,
315,
279,
5733,
2022,
304,
11113,
1124,
1097,
90,
5024,
24899,
2022,
27275,
641,
6588,
11,
4764,
11,
825,
3880,
311,
1896,
1119,
2692,
3807,
28720,
9363,
311,
198,
1782,
52741,
382,
5338,
... |
is composed of fewer materials than the total number of materials present in the image.
When sparsity is an a priori knowledge on the structure of the data, encouraging sparsity while computing NMF is likely to reduce noise and produce better results.
Sparse NMF is usually solved by extending standard NMF algorithms ... | 2006.07553 | https://arxiv.org/abs/2006.07553 | 2020-06-16 02:06:21 | 2006 | en | 4,096 | 2ac2642b-082f-4b02-be06-6ae860c1021c | 2,226 | [
374,
23415,
315,
16745,
7236,
1091,
279,
2790,
1372,
315,
7236,
3042,
304,
279,
2168,
624,
4498,
978,
70880,
374,
458,
264,
4867,
72,
6540,
389,
279,
5944,
315,
279,
821,
11,
25836,
978,
70880,
1393,
24231,
451,
31607,
374,
4363,
311,... |
symbol{X}}\quad& \text{rank}(\boldsymbol{X})\nonumber \\
\text{s.t.}\quad& \boldsymbol{y}_m=|\boldsymbol{A}_m
\boldsymbol{x}_m+\boldsymbol{w}_m| \quad \forall m \label{opt1}
\end{align}
where $\boldsymbol{y}_m\triangleq
[y_{1,m}\phantom{0}\ldots\phantom{0}y_{P,m}]^T$,
$\boldsymbol{w}_m\triangleq
[w_{1,m}\phantom... | 1811.01574 | https://arxiv.org/abs/1811.01574 | 2018-11-06 02:31:07 | 1811 | en | 4,096 | dbd781c7-882f-41c1-8b11-47b73433fb0c | 1,595 | [
18785,
90,
55,
3417,
59,
31610,
5,
1124,
1318,
90,
13129,
92,
11520,
14824,
18785,
90,
55,
5410,
59,
6280,
890,
24984,
319,
59,
1318,
84386,
734,
13,
11035,
31610,
5,
1124,
14824,
18785,
90,
88,
19444,
76,
28,
22957,
14824,
18785,
9... |
in order to achieve the best results.
The self-attention mechanism has proven successful on a variety of natural language processing tasks, but has not achieved ubiquitous success. The authors of \cite{kaiser2016} pointed out that an attention mechanism would likely struggle to solve a task which required a model to ... | 1912.11959 | https://arxiv.org/abs/1912.11959 | 2020-01-01 02:15:05 | 1912 | en | 4,096 | eb5c7c53-e6ff-465a-b7e2-3157a4a6c461 | 104 | [
304,
1973,
311,
11075,
279,
1850,
3059,
382,
785,
656,
12,
53103,
16953,
702,
16622,
6849,
389,
264,
8045,
315,
5810,
4128,
8692,
9079,
11,
714,
702,
537,
16994,
63464,
2393,
13,
576,
12014,
315,
1124,
67075,
69094,
39266,
17,
15,
16,... |
R$ samples from $\reallywidehat{p}$.
\end{description}
Step 1 involves the usual costs of learning real NVP networks, which can be done using gradient-based methods, such as ADAM \citep{ADAM}.
Assuming the number of layers and weights per layer in each network is fixed, evaluating $\reallywidehat{p}(\theta) = \prod_k... | 1903.04556 | https://arxiv.org/abs/1903.04556 | 2019-03-13 01:01:59 | 1903 | en | 4,096 | dceb40d4-8bd6-4696-9856-ea3186c6a931 | 4,828 | [
49,
3,
10469,
504,
57960,
53660,
9150,
9198,
90,
79,
92,
12947,
715,
59,
408,
90,
4684,
630,
8304,
220,
16,
17601,
279,
13475,
7049,
315,
6832,
1931,
451,
13378,
14155,
11,
892,
646,
387,
2814,
1667,
20169,
5980,
5413,
11,
1741,
438... |
}, Fig.~\ref{fig_experi_agent_env_lmrp2_cpc}) & 2.5 & 2.5 & 15.0 & 0.01\\
\hline
\end{tabular}
\end{center}
\end{table}
{\bf Training of PVED}. The proposed PVED model assumes paired $\left ( s_S, a_S \right )$ and $\left ( s_T, a_T \right )$ obtained from the training tasks. To generate paired trajectories, we train ... | 1908.05265 | https://arxiv.org/abs/1908.05265 | 2019-08-21 02:13:28 | 1908 | en | 4,096 | 261f59b7-de8f-42dc-aabb-fd3cbe1eae8d | 11,977 | [
2137,
23095,
13,
93,
59,
1097,
90,
904,
2702,
31736,
25730,
15879,
907,
20946,
79,
17,
666,
3992,
5410,
609,
220,
17,
13,
20,
609,
220,
17,
13,
20,
609,
220,
16,
20,
13,
15,
609,
220,
15,
13,
15,
16,
3422,
198,
59,
71,
1056,
... |
m(\beta)\neq 1$, and
$\f_0 = -\aprox(-\text{\upshape{C}}\star\beta + \tfrac{1}{2}\dotp{\alpha}{\text{\upshape{C}}\star\beta})$ otherwise.
\item \textbf{Power entropy:} $\f_0 = \rho(1-r)(m(\beta)^{\frac{1}{r - 1} } - 1) $, where $r$ is the dual exponent associated to $\phi^*$.
\end{itemize}
\subsection{Computing the ... | 1910.12958 | https://arxiv.org/abs/1910.12958 | 2021-03-22 01:12:38 | 1910 | en | 4,096 | 6d4e6ad9-eff5-452a-94ab-cbd217daa6f5 | 58,071 | [
76,
11520,
19127,
10699,
811,
80,
220,
16,
54876,
323,
198,
58736,
69,
62,
15,
284,
220,
481,
59,
391,
54789,
4080,
59,
1318,
35702,
454,
12231,
90,
34,
3417,
59,
11870,
59,
19127,
488,
1124,
83,
37018,
90,
16,
15170,
17,
11035,
1... |
loss $l(y, {\hat y}) = \max (0, 1-y \cdot {\hat y})$.
For each dataset, the hyperparameter $C$ was selected by $5$-fold cross validation among $10^{-2}, 10^{-1}, \dots, 10^{3}$
before enumeration.
We applied \algo{EnumSV} to these datasets, and enumerated top-$50$ models.
Figure~\ref{fig:1} presents the values of the... | 1906.01876 | https://arxiv.org/abs/1906.01876 | 2019-06-06 02:10:04 | 1906 | en | 4,096 | 122bb005-5505-4739-90bd-8aa14d89be40 | 4,725 | [
4709,
400,
75,
7021,
11,
28152,
9198,
379,
5410,
284,
1124,
2810,
320,
15,
11,
220,
16,
12034,
1124,
50853,
28152,
9198,
379,
5410,
3,
624,
2461,
1817,
10337,
11,
279,
17071,
16181,
400,
34,
3,
572,
4091,
553,
400,
20,
3,
12,
1996... |
_ssim}.
In Fig.~\ref{fig:semi_sup_quali}, we show reconstructed images obtained with such different levels.
It allows us to observe how image quality is affected when we gradually reduce the availability of labels.
Furthermore, we also evaluate the pose estimation accuracy with semi-supervision.
The overall PCK curves... | 1804.06364 | https://arxiv.org/abs/1804.06364 | 2020-02-18 02:01:30 | 1804 | en | 4,096 | 4bda17dd-24c4-45a5-b036-a0a9bed4542c | 32,228 | [
643,
14781,
27275,
641,
23095,
13,
93,
59,
1097,
90,
904,
25,
86598,
23723,
64391,
72,
2137,
582,
1473,
82004,
5335,
12180,
448,
1741,
2155,
5866,
624,
2132,
6147,
601,
311,
22986,
1246,
2168,
4271,
374,
11495,
979,
582,
26024,
7949,
... |
disturbances indeed depend on the patient's eating behaviour, which follows daily and weekly patterns. Similarly, parameters are typically subject to intra-patient variations such as daily oscillations. The parameter distribution $\Lambda(\lambda \mid t)$ can be alternatively defined to capture a population of patient... | 2003.01283 | https://arxiv.org/abs/2003.01283 | 2020-03-04 02:06:59 | 2003 | en | 4,096 | 3271b0f5-1c6c-425b-a6e5-5985948af10e | 11,960 | [
84060,
12824,
6761,
389,
279,
8720,
594,
12182,
16999,
11,
892,
11017,
7298,
323,
17059,
12624,
13,
34239,
11,
5029,
525,
11136,
3832,
311,
49838,
2268,
16389,
26244,
1741,
438,
7298,
42424,
804,
13,
576,
5733,
7982,
57960,
58266,
11520,
... |
150$ ms.
\subsubsection{Realistic Edit Tasks with Natural Adversarial Examples}
In all previous experiments, we considered edits with randomly chosen target class. However, in many practical scenarios, most of these edits will never occur. For instance, it is far more likely that an image previously classified... | 2004.00345 | https://arxiv.org/abs/2004.00345 | 2020-07-23 02:10:30 | 2004 | en | 4,096 | 636d9136-37ef-4256-bf85-5c20c6290466 | 3,090 | [
16,
20,
15,
3,
9829,
27427,
59,
1966,
68874,
90,
12768,
4532,
8340,
46471,
448,
18448,
2410,
3004,
42721,
25311,
3733,
641,
678,
3681,
21204,
11,
582,
6509,
50844,
448,
26618,
11882,
2169,
536,
13,
4354,
11,
304,
1657,
14976,
25283,
1... |
line{d}] p(c|\overline{a},\overline{d}) + E[Y|\overline{a},\overline{c},\overline{d}] p(\overline{c}|\overline{a},\overline{d})\\
&- E[Y|\overline{a},c,d] p(c|\overline{a},d) - E[Y|\overline{a},\overline{c},d] p(\overline{c}|\overline{a},d)\\
&= E[Y|\overline{a},c] (p(c|\overline{a},\overline{d}) - 0.5) + E[Y|\overline... | 2005.13245 | https://arxiv.org/abs/2005.13245 | 2020-08-25 02:16:05 | 2005 | en | 4,096 | db18eea1-249b-4285-aa4a-7c7021e4c698 | 5,904 | [
1056,
90,
67,
25439,
281,
1337,
22957,
1975,
1056,
90,
64,
2137,
59,
1975,
1056,
90,
67,
5410,
488,
468,
69907,
22957,
1975,
1056,
90,
64,
2137,
59,
1975,
1056,
90,
66,
2137,
59,
1975,
1056,
90,
67,
25439,
281,
11520,
1975,
1056,
... |
}} & 100 & 30 & \textbf{0.830} & 30, 100\\
endo & 0.848 & 0.839 & 0.848 & \underline{\textbf{0.851}} & 50 & 100 & 0.848 &10, 30, 100 \\
resp & 0.777 &0.703 & 0.770 & \underline{\textbf{0.784}} & 500 & 10 & \textbf{0.782} & 30 \\
inj & 0.662 & 0.590 & 0.653 & \textbf{0.683} & 500 & 10& \underline{\textbf{0.686}}... | 2004.00430 | https://arxiv.org/abs/2004.00430 | 2020-04-02 02:14:58 | 2004 | en | 4,096 | e40088d7-7e62-41cf-a63b-4a8f5baf2a7f | 9,445 | [
3417,
609,
220,
16,
15,
15,
609,
220,
18,
15,
220,
609,
1124,
1318,
13233,
90,
15,
13,
23,
18,
15,
92,
609,
220,
18,
15,
11,
220,
16,
15,
15,
3422,
715,
8691,
609,
220,
15,
13,
23,
19,
23,
609,
220,
15,
13,
23,
18,
24,
6... |
0$ & $1000$ & $0.995$ & $528$ & $1$ & $4$ \\
\bottomrule
\end{tabular}}
\end{table}
\subsection{Graph Normalizing Flow experiment}
The normalizing flows in the graph experiments all utilize three subflows, where a subflow consists of an actnorm layer \citep{kingma2018glow}, a $1 \times 1$ convolution and an a... | 2006.01910 | https://arxiv.org/abs/2006.01910 | 2020-10-27 01:39:01 | 2006 | en | 4,096 | 80a16f84-ef32-47dc-a98f-e7ad8e1d4a70 | 4,579 | [
15,
3,
609,
400,
16,
15,
15,
15,
3,
609,
400,
15,
13,
24,
24,
20,
3,
609,
400,
20,
17,
23,
3,
609,
400,
16,
3,
609,
400,
19,
3,
90155,
262,
1124,
14860,
12937,
198,
262,
1124,
408,
90,
6192,
1276,
11248,
59,
408,
90,
2005,... |
W_i(Y(X_{(i)})-1/2)>0 \bigg| X\in\mathcal{X}_{2p,\Delta}^-{\rm and } E^c \bigg)\\
&&+P\bigg( \sum_{i=1}^kW_i(Y(X_{(i)})-1/2)<0 \bigg| X\in\mathcal{X}_{2p,\Delta}^+{\rm and } E^c\bigg )\\
&\leq& B\bigg(\Delta+AC_1\bigg(\frac{k}{n}\bigg)^{\alpha/d}\bigg)^{\beta}+B\exp(-k/4)\\
&&+P\bigg( \sum_{i=1}^kW_i(Y(X_{(i)}... | 1810.02814 | https://arxiv.org/abs/1810.02814 | 2018-12-04 02:02:37 | 1810 | en | 4,096 | a49c25d9-659c-4036-8c7f-fb0cd6320f3e | 8,100 | [
54,
5318,
20206,
7644,
15159,
7,
72,
98329,
12,
16,
14,
17,
16018,
15,
1124,
16154,
70,
91,
1599,
59,
258,
59,
10374,
5416,
90,
55,
51535,
17,
79,
26266,
20277,
92,
49309,
35702,
8719,
323,
335,
468,
61,
66,
1124,
16154,
70,
8,
... |
and \underline{underlined} text refer to the 1st and 2nd ranked score, respectively.}
\label{tab:eyale_bench}
\end{table*}
\begin{figure*}[t]
\centering
\vspace{1cm}
\begin{subfigure}[b]{0.5\textwidth}
\centering
\includegraphics[width=\textwidth]{orl.png}
\caption{ORL}
... | 1908.09419 | https://arxiv.org/abs/1908.09419 | 2019-08-27 02:17:49 | 1908 | en | 4,096 | 135d9a4c-e00f-4697-896e-00d7e23ce346 | 8,628 | [
323,
1124,
45673,
90,
7995,
15121,
92,
1467,
8300,
311,
279,
220,
16,
267,
323,
220,
17,
303,
21006,
5456,
11,
15576,
13,
532,
59,
1502,
90,
6192,
25,
1195,
1574,
880,
19762,
532,
59,
408,
90,
2005,
9,
630,
59,
7265,
90,
17781,
... |
as $\frac{d}{dx}(u(x) v(x)) = \frac{du(x)}{dx}v(x) + u(x)\frac{dv(x)}{dx}$. This perfectly mechanistic process is realized in computer algebra systems such as Mathematica, Maple, and Maxima. Symbolic results can give insight into the problem and allow analytical solutions of optima (e.g. $\frac{df(x)}{dx}=0$) in which... | 1404.7456 | https://arxiv.org/abs/1404.7456 | 2014-04-30 02:11:45 | 1404 | en | 4,096 | 76d8fb53-4489-467e-9416-4de7e7c39488 | 951 | [
438,
57960,
37018,
90,
67,
15170,
12719,
25547,
84,
2075,
8,
348,
2075,
593,
284,
1124,
37018,
90,
1054,
2075,
9139,
90,
12719,
92,
85,
2075,
8,
488,
575,
2075,
10699,
37018,
90,
37261,
2075,
9139,
90,
12719,
92,
12947,
1096,
13942,
... |
de x_{wi}-x_{wi}) + \tilde x_i^Tx_wx_{wi}) \nonumber
\\ & + \frac{1}{\alpha_w^2} (x_i^Tx_wx_{wi}(\tilde x_{ww}-x_{ww}) + x_w^Tx_wx_{wi}\tilde x_{wi} + x_w^T\tilde x_w(x_{wi})^2 - x_w^Tx_w(x_{wi})^2) \nonumber
\\ & + \frac{1}{\alpha_w^3} x_w^Tx_w(x_{wi})^2(\tilde x_{ww}-x_{ww}) \label{eq:update_xtx}
\end{align}
\be... | 1804.04421 | https://arxiv.org/abs/1804.04421 | 2018-04-13 02:08:03 | 1804 | en | 4,096 | 9e4b82e6-c0a3-4d9c-a3c4-069f34740d8a | 4,391 | [
450,
856,
15159,
16632,
19732,
87,
15159,
16632,
5410,
220,
488,
1124,
1646,
450,
856,
5318,
61,
31584,
1670,
87,
15159,
16632,
5410,
1124,
6280,
890,
198,
220,
24984,
609,
488,
1124,
37018,
90,
16,
15170,
59,
7141,
1670,
61,
17,
92,
... |
& D_{\text{KL}}\big(\probtarget{\mathbf{x}}\| \probqphi{\mathbf{x}} \big) \\
& = \int \probtarget{\mathbf{x}} \log \left( \frac{\probtarget{\mathbf{x}}}{\probqphi{\mathbf{x}}}\right) \;d\mathbf{x}.
\end{align}
When minimizing it with respect to the parameters $\phi$ of the neural network, the objective ... | 2003.10200 | https://arxiv.org/abs/2003.10200 | 2020-03-24 01:26:46 | 2003 | en | 4,096 | 0a2e3945-8ab0-4d41-8b75-7538f5fd7336 | 4,107 | [
262,
609,
422,
15159,
59,
1318,
90,
52983,
3417,
59,
16154,
11520,
19899,
5657,
35702,
10374,
13233,
45340,
3417,
59,
91,
1124,
19899,
80,
16827,
35702,
10374,
13233,
45340,
3417,
1124,
16154,
8,
90155,
262,
609,
284,
1124,
396,
1124,
1... |
short}
\end{figure*}
\section{Formatting your paper}
All text must be in a two-column format. The total allowable width of the
text area is $6\frac78$ inches (17.5 cm) wide by $8\frac78$ inches (22.54
cm) high. Columns are to be $3\frac14$ inches (8.25 cm) wide, with a
$\frac{5}{16}$ inch (0.8 cm) space between them.... | 2004.08546 | https://arxiv.org/abs/2004.08546 | 2021-01-05 02:29:44 | 2004 | en | 4,096 | b3b19160-4fe2-412c-a2aa-4caf68ac249d | 3,697 | [
8676,
532,
59,
408,
90,
17781,
9,
630,
59,
2809,
90,
82135,
697,
5567,
630,
2403,
1467,
1969,
387,
304,
264,
1378,
24960,
3561,
13,
576,
2790,
84752,
2374,
315,
279,
198,
1318,
3082,
374,
400,
21,
59,
37018,
22,
23,
3,
14924,
320,... |
\sum_{t=0}^T \lambda_t \E \norm{\hat \xx_t - \xx_t}^2 \right)\,.
\end{align*}
Finally, using $\norm{\hat \xx_t - \xx_t}^2 = \delta_t \norm{\xx_t}^2$ by~\eqref{def:delta}, shows the theorem.
\end{proof}
Before giving the proof of Theorem~\ref{thm:nonconvex}, we first give a justification for the remark just below Theo... | 2006.07253 | https://arxiv.org/abs/2006.07253 | 2020-06-15 02:18:59 | 2006 | en | 4,096 | f1121a5d-f780-4ece-99a8-6f04a891964a | 14,092 | [
1124,
1242,
15159,
83,
28,
15,
92,
61,
51,
1124,
12935,
528,
1124,
36,
1124,
20011,
35702,
9198,
1124,
4146,
528,
481,
1124,
4146,
528,
92,
61,
17,
1124,
1291,
10699,
11,
624,
59,
408,
90,
6612,
9,
532,
23949,
11,
1667,
57960,
200... |
such matrices? Therefore, we introduce the following family of penalty functions:
\begin{definition}
Let $\mathbf{P} \in \mathbb{R}^{k \times p}$, all of its columns having unit norm. We introduce the norm $\Omega_{\mathbf{P}}$ as
\begin{equation*}
\Omega_{\mathbf{P}} (\mathbf{w}) = \| \mathbf{P} \Diag(\mathbf{w}) \|_... | 1109.1990 | https://arxiv.org/abs/1109.1990 | 2011-09-12 02:01:43 | 1109 | en | 4,096 | 66ddb606-4c7c-41da-bc35-a54e1e412f4c | 4,189 | [
1741,
35195,
30,
15277,
11,
582,
19131,
279,
2701,
2997,
315,
16346,
5746,
510,
59,
7265,
90,
18375,
532,
10061,
57960,
10374,
13233,
90,
47,
92,
1124,
258,
1124,
10374,
6066,
90,
49,
92,
47822,
74,
1124,
15136,
281,
31716,
11,
678,
... |
.pdf}
\vspace{-0.6 \baselineskip}
\caption{Illustration of an autoencoder for image compression.}
\vspace{-0.7 \baselineskip}
\label{fig:1}
\end{figure*}
The basic autoencoder training minimizes the image reconstruction error \cite{learning_deep_architectures}. In order to create a rate-distortion optimization, th... | 1802.09371 | https://arxiv.org/abs/1802.09371 | 2018-02-27 02:18:43 | 1802 | en | 4,096 | c59f0bdd-d165-49ed-9558-d6b7ef3c49f8 | 866 | [
15995,
532,
197,
72659,
8746,
19999,
15,
13,
21,
1124,
17797,
10999,
13389,
532,
197,
59,
23347,
90,
96157,
367,
315,
458,
3233,
27008,
369,
2168,
25111,
13,
532,
197,
72659,
8746,
19999,
15,
13,
22,
1124,
17797,
10999,
13389,
532,
19... |
vin algorithm in disordered systems have their root in the late 70s \cite{martin1973statistical,De78,kirkpatrick1987dynamics,CHS93}. Despite the dynamics was understood for some recurrent neural networks in long-time regime \cite{sompolinsky1988chaos,coolen2000statistical}, generalizing and solving the corresponding e... | 2001.00479 | https://arxiv.org/abs/2001.00479 | 2020-01-07 02:04:55 | 2001 | en | 4,096 | 32aa6008-97b3-4c96-9dbc-992f8310a9c0 | 1,223 | [
9603,
12111,
304,
827,
10544,
5942,
614,
862,
3704,
304,
279,
3309,
220,
22,
15,
82,
1124,
67075,
90,
33472,
258,
16,
24,
22,
18,
9878,
39496,
11,
1912,
22,
23,
16147,
14136,
64220,
16,
24,
23,
22,
67,
50148,
11,
2149,
50,
24,
1... |
{-0.5em}
\subsection{Single Opponent Type}\label{sec:single_opponent}
\vspace{-0.5em}
We now consider the special case where the learner is playing against the opponent of a single \emph{known} type at every round of the game, i.e., $\theta_t = \bar{\theta}$. The goal of the learner is to compete with the action that i... | 2007.05271 | https://arxiv.org/abs/2007.05271 | 2020-07-13 02:09:05 | 2007 | en | 4,096 | 8312fbea-d151-403b-94c7-3f4000c54a99 | 14,052 | [
19999,
15,
13,
20,
336,
532,
59,
68874,
90,
10888,
28610,
84547,
3990,
11035,
1502,
90,
5024,
14495,
2173,
10287,
1146,
532,
72659,
8746,
19999,
15,
13,
20,
336,
532,
1654,
1431,
2908,
279,
3281,
1142,
1380,
279,
62960,
374,
5619,
234... |
bf{\omega})^{3},~ G^{*}_{k}(\mathbf{\omega})=\sum_{i=1}^{r}\alpha_{k}^{*(i)}(\mathbf{v}_{k}^{*(i)}\cdot \mathbf{\omega})^{3}~~(k=1,2,\cdots,D)\]
Here, both $\mathbf{v}_{k}^{(i)}~(i=1,2,\cdots,r)$ and $\mathbf{v}_{k}^{*(i)}~(i=1,2,\cdots,r)$ are orthonormal vector groups, and the weights of the ground truth generator $\... | 2003.04033 | https://arxiv.org/abs/2003.04033 | 2021-02-19 02:02:43 | 2003 | en | 4,096 | 06838cc2-4416-4ce8-8d3c-2b4b6e87b0f2 | 2,166 | [
13233,
35702,
32696,
5410,
47822,
18,
2137,
93,
479,
47822,
9,
51535,
74,
92,
11520,
10374,
13233,
35702,
32696,
5410,
34433,
1242,
15159,
72,
28,
16,
92,
47822,
81,
11035,
7141,
15159,
74,
92,
47822,
6599,
72,
9139,
11520,
10374,
13233... |
exists policies under which $x'$ can be reached from $x$ with positive probability. For those MDPs, it is known that the \textit{average reward} following a stationary policy $\pi$, denoted by $\rho^{\pi}(\M)$ and defined as
\[
\rho^\pi(\M) = \lim\limits_{n\to \infty}\frac{1}{n} \E_{\M, \pi}\left(\sum_{t=0}^{n}... | 1004.5229 | https://arxiv.org/abs/1004.5229 | 2010-10-14 02:01:52 | 1004 | en | 4,096 | 4db37352-0062-4560-a9a7-cf81983dbd9a | 1,546 | [
6724,
10186,
1212,
892,
400,
87,
44701,
646,
387,
8643,
504,
400,
87,
3,
448,
6785,
18927,
13,
1752,
1846,
386,
10298,
82,
11,
432,
374,
3881,
429,
279,
1124,
1318,
275,
90,
17202,
11313,
92,
2701,
264,
52635,
4842,
57960,
2493,
548... |
text{(a)}}{\leq} &~ \frac{\sqrt{\E_{(s,a) \sim \mu}\left[\left(w_{\nicefrac{d_{\pi^\star}}{\mu}}(s,a)\right)^2\right]\E_{(s,a) \sim \mu}\left[\left((\Tcal \Qhat) (s,a) - \Qhat(s,a)\right)^2\right]}}{1 - \gamma} \\
&~ + \frac{\sqrt{\E_{(s,a) \sim \mu}\left[\left(w_{\nicefrac{d_{\pi_{\Qhat}}}{\mu}}(s,a)\right)^2\right]\E... | 2003.03924 | https://arxiv.org/abs/2003.03924 | 2020-08-25 02:23:22 | 2003 | en | 4,096 | 3b46cc13-7826-4379-8e36-625625900c30 | 45,723 | [
1318,
96065,
64,
52731,
35702,
273,
80,
92,
609,
93,
1124,
37018,
35702,
26888,
35702,
36,
15159,
7,
82,
15012,
8,
1124,
14781,
1124,
15128,
11035,
2359,
26056,
2359,
3622,
15159,
59,
52796,
37018,
90,
67,
15159,
59,
2493,
24884,
11870,... |
conditional density $p(\mathbf{x}\vert \mathbf{x}')=\mathcal{N}\left( \mu_{\mathbf{x}|\mathbf{x}'} , \Sigma_{\mathbf{x}|\mathbf{x}'}\right)$, and a pair of vectors $\mathbf{y},\mathbf{y}'$ such that $\mathbf{x}=\phi(\mathbf{y})$ and $\mathbf{x}'=\phi(\mathbf{y}')$, the conditional density $p(\mathbf{y}\vert\mathbf{y}'... | 1803.07102 | https://arxiv.org/abs/1803.07102 | 2018-03-21 01:00:32 | 1803 | en | 4,096 | db593178-ed71-4342-b948-094811229c76 | 477 | [
34687,
17457,
400,
79,
11520,
10374,
13233,
45340,
11035,
1621,
1124,
10374,
13233,
45340,
67183,
34433,
10374,
5416,
90,
45,
11035,
2359,
7,
1124,
15128,
15159,
59,
10374,
13233,
45340,
92,
22957,
10374,
13233,
45340,
92,
8275,
1154,
1124,... |
with $n$ nodes in total, $\mathcal{E}$ is the edge set, and $\mathbf{X}=[\mathbf{x}_1, \mathbf{x}_2. \cdots, \mathbf{x}_n]^T$ is the feature matrix. The topology structure of graph $\mathcal{G}$ can be denoted by an adjacency matrix $\mathbf{A}=\{a_{ij}\} \in \mathbb{R}^{n\times n}$, where $a_{ij}=1$ if $(v_i,v_j)\in ... | 2007.01594 | https://arxiv.org/abs/2007.01594 | 2020-07-06 02:11:43 | 2007 | en | 4,096 | bfb42ceb-5bf4-499d-ba34-748921b8a67e | 2,209 | [
448,
400,
77,
3,
7798,
304,
2790,
11,
57960,
10374,
5416,
90,
36,
31716,
374,
279,
6821,
738,
11,
323,
57960,
10374,
13233,
90,
55,
92,
5818,
59,
10374,
13233,
45340,
19444,
16,
11,
1124,
10374,
13233,
45340,
19444,
17,
13,
1124,
43... |
stationary and non-stationary signals. However, it can face difficulties in decomposing the signal into unique frequency components, resulting in mode mixing~\cite{huang2014hilbert}.
\begin{figure}[t]
\centering
\subfloat[Beginning of training (motion).]{
\includegraphics[width=0.24\textwidth]{./tsne_noisyTrai... | 1812.08555 | https://arxiv.org/abs/1812.08555 | 2020-07-07 02:16:27 | 1812 | en | 4,096 | 8f0274c3-84ee-4f50-b88a-ac4073e60931 | 356 | [
52635,
323,
2477,
5477,
367,
658,
17289,
13,
4354,
11,
432,
646,
3579,
26038,
304,
28502,
8314,
279,
8286,
1119,
4911,
11639,
6813,
11,
12942,
304,
3856,
26792,
93,
59,
67075,
90,
16739,
524,
17,
15,
16,
19,
75323,
9160,
7810,
4710,
... |
{Lagrangian Mechanics}%
Describing the trajectory of a system has been extensively studied, and various mathematical
formulations exist to derive the corresponding differential equations. Within this approach, a modified form of the Lagrange mechanics is used, i.\,e., the Euler--Lagrange formulation with generalized co... | 2005.14617 | https://arxiv.org/abs/2005.14617 | 2020-06-01 02:13:49 | 2005 | en | 4,096 | 0e3ff3af-c302-45a1-a23f-4fa2d89a50e5 | 872 | [
90,
43,
68305,
524,
1103,
76823,
92,
13744,
4896,
43531,
279,
34682,
315,
264,
1849,
702,
1012,
41717,
19476,
11,
323,
5257,
35972,
198,
627,
7455,
3000,
311,
42430,
279,
12159,
40164,
37906,
13,
24236,
419,
5486,
11,
264,
10807,
1352,
... |
mathcal{x})}'{\m{R}}_{\s{A}}^{-1}\big(\vg{\gamma}-\m{G}_{\s{A}}\hat{\vg{\beta}}\big),\label{fity0}\end{eqnarray}
where ${\mathbf}\def\m{\mathbf}\def\vg{\boldsymbol}\def\s{\mathcal{r}}_{\s{A}}$ and ${\m{R}}_{\s{A}}$ are the same as in \eqref{ki}, $\hat{\vg{\beta}}=(\m{G}_{\s{A}}'\m{R}_{\s{A}}^{-1}\m{G}_{\s{A}})^{-1}\m{G... | 1805.10122 | https://arxiv.org/abs/1805.10122 | 2019-11-28 02:13:50 | 1805 | en | 4,096 | 1bcf2e83-8016-4be4-915f-0dc8d772fcf3 | 9,274 | [
10374,
5416,
45340,
5410,
11661,
35702,
76,
90,
49,
3417,
15159,
59,
82,
90,
32,
3417,
87210,
16,
11035,
16154,
11520,
7239,
35702,
32214,
19732,
59,
76,
90,
38,
51535,
59,
82,
90,
32,
3417,
59,
9198,
35702,
7239,
35702,
19127,
3417,
... |
more complex Mujoco tasks, exploration and learning must be interleaved. We combine model predictive control (MPC) \citep{mayne2000constrained} with the actor-critic method (DDPG) for planning. MPC refines the stochastic model-free policy (the actor) by sampling-based planning \citep{wang2019exploring, hong2019model},... | 2006.16210 | https://arxiv.org/abs/2006.16210 | 2020-10-27 01:22:30 | 2006 | en | 4,096 | 3097a800-0f53-4c5b-b91e-9174c4825bbc | 10,644 | [
803,
6351,
57236,
13140,
9079,
11,
26403,
323,
6832,
1969,
387,
94968,
4141,
13,
1205,
15963,
1614,
59236,
2524,
320,
44,
4872,
8,
1124,
67075,
79,
90,
18358,
811,
17,
15,
15,
15,
443,
57727,
92,
448,
279,
12089,
1786,
49208,
1714,
... |
" \cite{blasiusinterview}, especially due to their limited manpower. This limited manpower meant that such a solution would also have to be very precise, with one investigator stating they could not tolerate more than 50\% of the safety issues identified being false positives \cite{cpscinterview2}.
Domain literature s... | 1805.09772 | https://arxiv.org/abs/1805.09772 | 2018-07-24 02:09:03 | 1805 | en | 4,096 | 63f8bc69-fac4-4905-a2fd-f8c70561b50c | 550 | [
1,
1124,
67075,
90,
49297,
9156,
396,
12873,
2137,
5310,
4152,
311,
862,
7199,
95769,
13,
1096,
7199,
95769,
8791,
429,
1741,
264,
6291,
1035,
1083,
614,
311,
387,
1602,
23560,
11,
448,
825,
48481,
27798,
807,
1410,
537,
49034,
803,
1... |
summarizes our off-policy methodology. For on-policy, Eq. \ref{eq: on-policy} can be used for gradient estimates to avoid using an extra model $p_{\theta'}$. One can use the difference between $p_{\theta^ {(t)}}(x)$ and $p_{\theta^{(t-t_0)}}(x)$ for some $t_0$ being less than some threshold as the stopping criteria. D... | 2002.07236 | https://arxiv.org/abs/2002.07236 | 2020-02-19 02:02:00 | 2002 | en | 4,096 | 95bde9b4-494c-431f-a27e-875dd7dd47c2 | 4,387 | [
93494,
1039,
1007,
66420,
37052,
13,
1752,
389,
66420,
11,
33122,
13,
1124,
1097,
90,
11006,
25,
389,
66420,
92,
646,
387,
1483,
369,
20169,
17530,
311,
5648,
1667,
458,
4960,
1614,
400,
79,
15159,
59,
15976,
8275,
12947,
3776,
646,
9... |
c}
\hline
Environment & State Dim & Action Dim & Reward threshold & Expert Performance \\
\hline
Swimmer-v2 & 8 & 2 & 360 & 332\\
\hline
Hopper-v2 & 11 & 3 & 3800 & 3566\\
\hline
Walker2d-v2 & 17 & 6 & - & 4924\\
\hline
Ant-v2 & 111 & 8 & 6000 & 6157\\
\hline
HalfCheetah-v2 & 17 & 6 & 4800 & 12294\\
\hline
Humanoid-v2 ... | 1911.10947 | https://arxiv.org/abs/1911.10947 | 2019-11-26 02:30:40 | 1911 | en | 4,096 | af162a6f-0421-4385-a330-c60d5d98938f | 12,666 | [
66,
532,
59,
71,
1056,
198,
12723,
609,
3234,
8126,
609,
5586,
8126,
609,
55800,
12171,
609,
32157,
20651,
90155,
59,
71,
1056,
198,
13218,
18982,
8273,
17,
609,
220,
23,
609,
220,
17,
609,
220,
18,
21,
15,
609,
220,
18,
18,
17,
... |
0)-\mathcal{F}(\hat{z}_1,\ldots,\hat{z}_{i-1},0,0,\ldots,0)\nonumber\\
& =g(\hat{z}_i)-g(0)=g(\hat{z}_i)+\mathcal{F}({Z}_{l},\mathbf{X}_{-i}^{(i-1)})-\mathcal{F}(\mathbf{X}^{(i-1)})\nonumber\\
&\geq g(\hat{z}_i)+\mathcal{F}({Z}_{l},\mathbf{Y}_{-i}^{(i-1)})-\mathcal{F}(0,\mathbf{Y}_{-i}^{(i-1)})\geq g(\hat{z}_i)\label{e... | 1805.09480 | https://arxiv.org/abs/1805.09480 | 2018-05-25 02:04:23 | 1805 | en | 4,096 | 77f8063f-48c6-4f12-90fe-68ade48e8012 | 17,851 | [
15,
7287,
59,
10374,
5416,
90,
37,
92,
11520,
9198,
90,
89,
19444,
16,
26266,
507,
2412,
26266,
9198,
90,
89,
51535,
72,
12,
16,
2137,
15,
11,
15,
26266,
507,
2412,
11,
15,
10699,
6280,
890,
3422,
198,
5,
284,
70,
11520,
9198,
9... |
umption}
\label{as:expected_smoothness}
The function $f$ is $\mathcal{L}-$smooth with respect to a datasets $\mathcal{D}$ if there exist $\mathcal{L} > 0$ with
\begin{equation}\label{eq:exp_smoothn_cL}
\mathbb{E}_\mathcal{D}\left[\left\| \nabla f_v(x) - \nabla f_v(x^*) \right\|^2\right] \leq 2 \mathcal{L}(f(x) - f(x^*)... | 2005.01097 | https://arxiv.org/abs/2005.01097 | 2021-11-22 02:21:26 | 2005 | en | 4,096 | 87d7f658-563c-4949-87ec-cd84c21dcb7f | 13,813 | [
60574,
532,
59,
1502,
90,
300,
25,
7325,
71655,
2090,
532,
785,
729,
400,
69,
3,
374,
57960,
10374,
5416,
90,
43,
19732,
3,
56866,
448,
5091,
311,
264,
29425,
57960,
10374,
5416,
90,
35,
31716,
421,
1052,
3000,
57960,
10374,
5416,
9... |
=80mm]{sales_v2.png}}
\fbox{\includegraphics[height=50mm, width=80mm]{mixture_v2.png}}
\begin{tabular}{p{8cm}p{8cm}}
\centering (a) & \centering (b) \\
\end{tabular}
\caption{(a) An example time-series of the demand for a particular SKU. One can observe the large number of peaks and valleys in the... | 1803.03800 | https://arxiv.org/abs/1803.03800 | 2018-03-19 01:05:15 | 1803 | en | 4,096 | 89a18647-e2f9-4326-aaa0-c2e67afc949f | 5,937 | [
28,
23,
15,
3821,
15370,
29041,
2273,
17,
3508,
47449,
5872,
262,
1124,
69,
2011,
35702,
997,
31685,
58,
2563,
28,
20,
15,
3821,
11,
2374,
28,
23,
15,
3821,
15370,
76,
12735,
2273,
17,
3508,
11248,
262,
1124,
7265,
90,
6192,
1276,
... |
max & 10 units \\
\bottomrule
\end{tabular}
\end{center}
\label{MNIST_struc}
\end{table}
Once an ensemble of classifiers is trained, it predicts by letting each
classifier vote for a label.
More specifically, the predicted value is chosen to be the label that maximizes
the average of the output probabilities ... | 1709.03423 | https://arxiv.org/abs/1709.03423 | 2018-02-09 02:04:46 | 1709 | en | 4,096 | 946b00e7-2210-4461-95d7-7ea245a848eb | 4,717 | [
2810,
609,
220,
16,
15,
8153,
24984,
715,
59,
14860,
12937,
198,
262,
1124,
408,
90,
6192,
1276,
532,
59,
408,
90,
3057,
532,
59,
1502,
90,
55181,
3846,
2895,
1754,
532,
59,
408,
90,
2005,
3733,
12522,
458,
39026,
315,
71291,
374,
... |
ample2016}, which goes back to the works of \cite{Chiu2016NamedER,Huang2015BidirectionalLM,collobert2011natural}. We use a neural model consisting of stacked LSTM and CRF layers. The \textit{base layer} is made of two parts: (i) a preprocessing sublayer generating the character-based embeddings with a cell of forward a... | 1807.10675 | https://arxiv.org/abs/1807.10675 | 2018-07-30 02:11:47 | 1807 | en | 4,096 | 2854ff26-c8e4-4686-9d8f-45f478989bbb | 1,964 | [
1516,
17,
15,
16,
21,
2137,
892,
5780,
1182,
311,
279,
4278,
315,
1124,
67075,
90,
1143,
18738,
17,
15,
16,
21,
15810,
640,
43539,
68610,
17,
15,
16,
20,
65452,
44670,
10994,
51496,
1684,
529,
17,
15,
16,
16,
52880,
7810,
1205,
99... |
conservation of unbalancedness magnitude throughout optimization (Subappendix~\ref{app:unbalanced:conserve}).
Both approaches rely on the fact that small unbalancedness magnitude implies proximity to perfect balancedness, as stated formally in the following lemma.
\begin{lemma} \label{lem:unbalance_to_balance}
For any... | 2005.06398 | https://arxiv.org/abs/2005.06398 | 2020-10-20 02:13:09 | 2005 | en | 4,096 | 4d2f78f4-87d6-4399-ad09-50069ff997c1 | 69,162 | [
28611,
315,
650,
58402,
2090,
25621,
6814,
25262,
320,
3136,
5090,
941,
93,
59,
1097,
90,
676,
25,
359,
58402,
25,
443,
13267,
92,
4292,
20629,
19827,
17188,
389,
279,
2097,
429,
2613,
650,
58402,
2090,
25621,
23945,
36743,
311,
4727,
... |
Nearby' ring of subclasses for `nearby N' (`nN'), `nearby NE' (`nNE'), etc.
Suppose `nN' is designated as a template subclass with some set of nominal (but unknown) desired CLOP normal vectors and biases $\left\{n_{nN,NT},n_{nN,fN},n_{nN,nNE},n_{nN,nNW}\right\}$ (`NT': `Next To', `fN': `far North').
Rotational symmet... | 1806.00728 | https://arxiv.org/abs/1806.00728 | 2018-07-24 02:09:34 | 1806 | en | 4,096 | fa8933cf-46e0-45a2-84fd-81c433b35795 | 6,319 | [
51962,
1694,
6,
10058,
315,
68183,
369,
1565,
51659,
1694,
451,
6,
28654,
77,
45,
4567,
1565,
51659,
1694,
7856,
6,
28654,
77,
3944,
4567,
4992,
13,
715,
10048,
2900,
1565,
77,
45,
6,
374,
23195,
438,
264,
3811,
37190,
448,
1045,
73... |
).\footnote{A cache set can have less than $W$ cache lines for the first $W-1$ cache accesses (small fraction of program execution). In this case, no eviction is needed to insert the line.}
\item $s^h_t = (\{m_1, \dots, m_{t-1}\}, \{pc_1, \dots, pc_{t-1}\})$ is the history of all past cache accesses.
In pr... | 2006.16239 | https://arxiv.org/abs/2006.16239 | 2020-07-13 02:03:04 | 2006 | en | 4,096 | 8ca6ad96-aafb-4df6-9db7-f307d3abee0e | 3,095 | [
72341,
5334,
9974,
90,
32,
6500,
738,
646,
614,
2686,
1091,
400,
54,
3,
6500,
5128,
369,
279,
1156,
400,
54,
12,
16,
3,
6500,
77648,
320,
9004,
19419,
315,
2025,
11320,
568,
758,
419,
1142,
11,
902,
78236,
374,
4362,
311,
5656,
27... |
with each class using offline training data for clinically stable patients. Discovering the latent classes is carried out using unsupervised learning over the domain of clinically stable patients since they are dominant in the dataset, and are more likely to exhibit stationary physiological trajectories, thus their ph... | 1605.00959 | https://arxiv.org/abs/1605.00959 | 2016-05-04 02:12:42 | 1605 | en | 4,096 | dc828d49-6561-4bb6-a979-9f7cbe59ec0a | 735 | [
448,
1817,
536,
1667,
26166,
4862,
821,
369,
69332,
15175,
6835,
13,
32939,
287,
279,
41667,
6846,
374,
11691,
700,
1667,
6975,
454,
77990,
6832,
916,
279,
7947,
315,
69332,
15175,
6835,
2474,
807,
525,
24456,
304,
279,
10337,
11,
323,
... |
p_{{{\boldsymbol \theta}}}(\boldsymbol{x})_{-}}=\emptyset$. By contrast, there is not a substantial penalty to
$\mathcal{S}_{q(\boldsymbol{x})_{-}}\cap \mathcal{S}_{p_{{{\boldsymbol \theta}}}(\boldsymbol{x})}\neq\emptyset$.
Summarizing these conditions, the goal of maximizing $-\mbox{KL}(q(\boldsymbol{x})\|p_{{\... | 1709.01846 | https://arxiv.org/abs/1709.01846 | 2017-10-23 02:03:17 | 1709 | en | 4,096 | c9915038-38ea-416f-83d8-e705c4c2ba01 | 3,999 | [
79,
62,
2979,
35702,
14824,
18785,
1124,
15976,
75542,
11520,
14824,
18785,
45340,
5410,
15159,
12,
3417,
34433,
3194,
746,
12947,
3216,
12872,
11,
1052,
374,
537,
264,
11924,
16346,
311,
2303,
197,
58736,
10374,
5416,
90,
50,
51535,
80,
... |
widebar{C}^2}{\epsilon^2}\right)^{\textnormal{\texttt{dim}}_\mathrm{F}}\cdot\exp{\left(-\frac{n\epsilon^2}{48\cdot214\widebar{C}^4}\right)} \nonumber \\
&\qquad +\exp{\left(-\frac{3}{416}\cdot\frac{n\epsilon^2}{\widebar{C}^2}\right)} \nonumber
\end{align}
Note that $\inf_{f\in\mathrm{F}}\norm{f-\mathbb{T} Q_{k-1}}... | 1903.08738 | https://arxiv.org/abs/1903.08738 | 2019-03-22 01:03:07 | 1903 | en | 4,096 | 32e53ec7-342a-4417-a12c-f4d912b2a484 | 15,438 | [
9150,
2257,
90,
34,
92,
61,
17,
15170,
59,
31767,
61,
17,
11035,
1291,
29776,
35702,
1318,
8252,
35702,
1318,
5566,
90,
12927,
3417,
56014,
91550,
90,
37,
3417,
59,
50853,
59,
4580,
35702,
2359,
4080,
59,
37018,
91362,
59,
31767,
61,
... |
2\beta \ell}{17A^8\gamma \log n}.$$
\paragraph{Final steps.}
Now, combining all of the union bounds and conditions from \eqref{uniona}, \eqref{unionb}, \eqref{unionc}, \eqref{uniond} and \eqref{unione}, assuming that $\beta, \gamma = \Theta(1)$ and that $\omega(1) \leq \ell(n) \leq o(n/\log n)$, under the condition
\b... | 1907.11635 | https://arxiv.org/abs/1907.11635 | 2020-03-03 02:23:16 | 1907 | en | 4,096 | 510fbd21-aa63-470e-9944-ff904bbea235 | 21,316 | [
17,
59,
19127,
1124,
613,
15170,
16,
22,
32,
61,
23,
59,
32214,
1124,
839,
308,
92,
2418,
66426,
59,
27727,
90,
19357,
7354,
13,
532,
7039,
11,
34171,
678,
315,
279,
11300,
14262,
323,
4682,
504,
1124,
11006,
1097,
90,
16192,
64,
... |
tasks.
\begin{enumerate}
\item Exact Semantic Map\\
When an exact equivalence between a user supplied model(referred to as primal model), and it's neural network counterpart(referred to as proxy model) exists, we can initialize the weights of the neural network with the parameters of the fitted model, and just save ... | 2001.00818 | https://arxiv.org/abs/2001.00818 | 2020-01-06 02:07:02 | 2001 | en | 4,096 | e9773613-d2ba-4547-b9de-e50313116aeb | 2,865 | [
9079,
382,
59,
7265,
90,
76569,
630,
59,
1203,
74480,
74333,
5027,
3422,
198,
4498,
458,
4734,
84162,
1948,
264,
1196,
17221,
1614,
5801,
5554,
311,
438,
86181,
1614,
701,
323,
432,
594,
29728,
3922,
44016,
5801,
5554,
311,
438,
13291,
... |
{\nabla q(\mathbf{h})}{\mathbf{h}'-\mathbf{h}} + \frac{1}{2\eta} \norm{\mathbf{h}'-\mathbf{h}}_2^2,\nn
\end{equation}
$q(\mathbf{h}) \triangleq \frac{1}{2}\norm{\mathbf{v}-\mathbf{W}\mathbf{h}-\mathbf{r}}_2^2$ and $\eta\in\left(0,1/L\right]$ with $L\triangleq \norm{\mathbf{W}}_2^2$. The steps \eqref{eq:min_h} and \eqr... | 1604.02634 | https://arxiv.org/abs/1604.02634 | 2016-10-18 02:03:33 | 1604 | en | 4,096 | aabdde9e-2194-4e59-859f-03e845dabdf6 | 7,179 | [
35702,
77,
370,
4260,
2804,
11520,
10374,
13233,
90,
71,
5410,
15170,
59,
10374,
13233,
90,
71,
11661,
30529,
10374,
13233,
90,
71,
3417,
488,
1124,
37018,
90,
16,
15170,
17,
59,
1915,
92,
1124,
20011,
35702,
10374,
13233,
90,
71,
116... |
we can simulate $(C_n,M_n,\theta_n)$ from (our approximation to) $p(C_n,M_n,\theta_n|y_{1:n})$. These will give us the start of the final segment, together with its model and parameter values. Assume we simulate $C_n=t$, then we will next simulate $(C_t,M_t,\theta_t)$ from
\[
p(C_t,M_t,\theta_t|y_{1:n},C_{t+1}=t,C_{t... | 0910.3099 | https://arxiv.org/abs/0910.3099 | 2009-10-16 14:22:40 | 0910 | en | 4,096 | 0643bf3e-1719-43f4-bcaa-07a070cc8a9a | 6,673 | [
582,
646,
37453,
4930,
34,
1089,
27014,
1089,
26266,
15976,
1089,
15087,
504,
320,
413,
56204,
311,
8,
400,
79,
3025,
1089,
27014,
1089,
26266,
15976,
1089,
91,
88,
15159,
16,
19266,
5410,
12947,
4220,
686,
2968,
601,
279,
1191,
315,
... |
second moments. Then $ h_{ij} = H_{ij}( \Xi_{i,j-1} ) $ for
functions $ H_{ij} $. Obviously, $ \varepsilon_{ij} $ is $ \mathcal{F}_{ij}
$-measureable and, since $ \xi_{ij} $ is independent from the random
variables of the set $ \Xi_{i,j-1} $, we have $ E( \xi_{ij} |
\mathcal{F}_{i,j-1} ) = E(\xi_{ij}) = 0 $ yielding
\... | 1001.1841 | https://arxiv.org/abs/1001.1841 | 2010-01-12 11:43:51 | 1001 | en | 4,096 | 4317fa47-75ba-40e2-a353-8416e92ef814 | 7,754 | [
2086,
13943,
13,
5005,
400,
305,
15159,
3172,
92,
284,
472,
15159,
3172,
25547,
1124,
70630,
15159,
72,
10327,
12,
16,
92,
873,
400,
369,
198,
21409,
400,
472,
15159,
3172,
92,
8672,
35430,
11,
400,
1124,
85,
546,
59892,
15159,
3172,
... |
}} \E \{ |Y - \beta_f^T Z(W_{-k}) - \beta_k z(w_k)|^2 \} + \kappa ( \| \beta_f \|_1 + |\beta_k| )~, $$
where $W_{-k}$ is a shorthand for the matrix containing the rest of the vectors that have not been discarded yet.
Removing the vector $w_k$ from the first layer
increases the loss by a factor that is upper bounded ... | 1611.01540 | https://arxiv.org/abs/1611.01540 | 2017-06-05 02:01:52 | 1611 | en | 4,096 | ba96a380-c781-45fd-bd04-fb3bf08fa4c7 | 13,966 | [
3417,
1124,
36,
1124,
90,
760,
56,
481,
1124,
19127,
761,
61,
51,
1863,
14031,
15159,
12,
74,
5410,
481,
1124,
19127,
4698,
1147,
3622,
4698,
17935,
61,
17,
1124,
92,
488,
1124,
74,
27180,
320,
97155,
1124,
19127,
761,
97155,
62,
16... |
that $ad-bc\neq 0$.
These transformations are called {\em projective linear} or {\em M\"obius transformations}, and form the group $\PGL_2(\mathbb{Q}_p)$. The reason for
invariance under $\PGL_2(\mathbb{Q}_p)$ is the well known fact that M\"obius
transformations take equivalent disks to equivalent disks.
\medskip
As ... | 0707.3536 | https://arxiv.org/abs/0707.3536 | 2007-07-24 14:34:33 | 0707 | en | 4,096 | fb79c4d5-ff3e-4fb3-83e6-8eb021b4972b | 3,278 | [
429,
400,
329,
1455,
66,
59,
811,
80,
220,
15,
3,
624,
9485,
52970,
525,
2598,
28152,
336,
2390,
533,
13482,
92,
476,
28152,
336,
386,
2105,
674,
9156,
52970,
2137,
323,
1352,
279,
1874,
57960,
47,
3825,
62,
17,
11520,
10374,
6066,
... |
in $S_{add}$, we denote the first and the second highest values of $F(\mathbf{x})$ by $h_1$ and $h_2$, and then set
$Dis_\mathbf{x} = h_1-h_2$. We take $k$ samples with smallest $Dis_\mathbf{x}$ as $S_{add}$.
\end{itemize}
Random Select method is considered as the baseline method. Max Entropy method uses inf... | 1809.04913 | https://arxiv.org/abs/1809.04913 | 2018-09-14 02:08:53 | 1809 | en | 4,096 | 1ea8cf39-c65f-484f-9f93-2f15e9235e2b | 5,415 | [
304,
400,
50,
15159,
718,
31716,
11,
582,
78064,
279,
1156,
323,
279,
2086,
8426,
2750,
315,
400,
37,
11520,
10374,
13233,
45340,
5410,
3,
553,
400,
71,
62,
16,
3,
323,
400,
71,
62,
17,
54876,
323,
1221,
738,
715,
262,
400,
4839,
... |
6n_s) + 2\log(8/\delta)}{n_s}}.
\end{align}
Recall that
\begin{align*}
\hat{\epsilon}_{\text{split}}
= \max_{s\in \{0,1\}} L_s(\hat{h}^*) - \max_{s\in \{0,1\}} L_s(\hat{h}^*_s).
\end{align*}
Now by \eqref{eq::samp_comp_dif}, we conclude that
\begin{align*}
\left|\hat{\epsilon}_{\text{split}} - \hat{\epsi... | 2002.04788 | https://arxiv.org/abs/2002.04788 | 2020-02-13 02:05:53 | 2002 | en | 4,096 | af79c733-7c1a-485e-8e00-288fb503b8c9 | 6,718 | [
21,
77,
643,
8,
488,
220,
17,
59,
839,
7,
23,
34319,
20255,
9139,
90,
77,
643,
3417,
624,
59,
408,
90,
6612,
532,
3820,
541,
429,
715,
59,
7265,
90,
6612,
9,
532,
262,
1124,
9198,
35702,
31767,
51535,
59,
1318,
90,
6960,
3417,
... |
}) by defining $p_{i-1}(m) = P(m|M_i,I)\, P(d|m,M_i,I)$ ($i=1,2$) and sampling from the following family of bridging distributions
\begin{equation}
p(x|\beta) \propto [P(m|M_1,I)\, P(d|m,M_1,I)]^{1-\beta}\, [P(m|M_2,I)\, P(d|m,M_2,I)]^{\beta}
\end{equation}
For the special case that both models also share the same ... | 1411.3013 | https://arxiv.org/abs/1411.3013 | 2015-11-24 02:14:48 | 1411 | en | 4,096 | 9b0beb10-4f33-477c-a5d4-2f1011102ca7 | 8,795 | [
5410,
553,
26312,
400,
79,
15159,
72,
12,
16,
25547,
76,
8,
284,
393,
1255,
86878,
5318,
37768,
10699,
11,
393,
1500,
90563,
27014,
5318,
37768,
15087,
1711,
72,
28,
16,
11,
17,
3,
8,
323,
24892,
504,
279,
2701,
2997,
315,
76747,
... |
ary matrix and $\Lambda$ is a diagonal matrix. Since all the entries of $\Lambda$ are non-negative (in fact strictly positive), we can write $A = B*B$ where $B = Q\Lambda^{1/2}$.
Therefore,
$\mbox{det}(A) = |\mbox{det}(B)|^2 \geq 0$. Further $\mbox{det}(A) \ne 0$, since other $\mbox{det}(A - 0.I) = 0$ i.e. 0 is an ei... | 2005.01566 | https://arxiv.org/abs/2005.01566 | 2020-05-05 02:32:14 | 2005 | en | 4,096 | 5039568f-4814-4ea2-b098-80150ae6d2e7 | 12,146 | [
658,
6172,
323,
57960,
58266,
3,
374,
264,
37236,
6172,
13,
8704,
678,
279,
10695,
315,
57960,
58266,
3,
525,
2477,
60935,
320,
258,
2097,
25470,
6785,
701,
582,
646,
3270,
400,
32,
284,
425,
81339,
3,
1380,
400,
33,
284,
1207,
59,
... |
$ and $m = 3$, we have
$V(S_1) = \{ v_1,v_2 \}$ and $V(S_3) = \{ v_3,v_1\}$.
These $k$-node subgraphs are connected in $G^{(k)}$
whenever they share $k-1$ nodes between them, according to Definition~\ref{def:hon}, i.e.,
$S_i$ is connected to $S_{i-1}$ and $S_{i+1}$ in $G^{(k)}$, $\forall i \in \{ 2 \leq i \leq m-1 \}$.... | 1809.05241 | https://arxiv.org/abs/1809.05241 | 2018-09-17 02:05:29 | 1809 | en | 4,096 | 75ad1dd9-1898-4b92-ac05-27b4e2193a2d | 13,311 | [
3,
323,
400,
76,
284,
220,
18,
54876,
582,
614,
198,
3,
53,
3759,
62,
16,
8,
284,
1124,
90,
348,
62,
16,
14270,
62,
17,
1124,
31716,
323,
400,
53,
3759,
62,
18,
8,
284,
1124,
90,
348,
62,
18,
14270,
62,
16,
59,
31716,
624,
... |
17} that eigenfunctions of transfer operators and embedded transfer operators can be estimated simply by solving an eigenvalue problem involving (time-lagged) Gram matrices. The resulting methods are closely related to kernel Extended Dynamic Mode Decomposition (EDMD) and its variants \cite{WRK15, SP15}. As a result, i... | 1805.10118 | https://arxiv.org/abs/1805.10118 | 2018-05-28 02:07:42 | 1805 | en | 4,096 | d388cbbf-62f7-41cc-9d71-b122a27c7c18 | 302 | [
16,
22,
92,
429,
28724,
21409,
315,
8317,
19624,
323,
22864,
8317,
19624,
646,
387,
12943,
4936,
553,
21828,
458,
28724,
957,
3491,
15860,
320,
1678,
2852,
96476,
8,
28837,
35195,
13,
576,
12942,
5413,
525,
15148,
5435,
311,
10001,
4056... |
QN & ReduMIS & Ours\\
\toprule
ego-Facebook & 993 & 1,020 & \bfseries 1,046 & \bfseries 1,046 & 3,046 & 3,019 & \bfseries 2,993 & \bfseries 2,993\\
ego-Gplus & 56,866 & 56,603 & \bfseries 57,394 & \bfseries 57,394 & 50,748 & 51,011 & \bfseries 50,220 & \bfseries 50,220\\
ego-Twitter & 36,235 & 36,275 & \bfserie... | 1810.10659 | https://arxiv.org/abs/1810.10659 | 2018-10-26 02:04:27 | 1810 | en | 4,096 | 18464d5a-27d6-4c04-b356-794c1ff8193e | 5,547 | [
87054,
609,
3731,
84,
44,
1637,
609,
506,
1723,
3422,
198,
220,
1124,
3481,
12937,
198,
220,
36274,
7276,
4497,
609,
220,
24,
24,
18,
609,
220,
16,
11,
15,
17,
15,
609,
1124,
13233,
19880,
220,
16,
11,
15,
19,
21,
609,
1124,
132... |
}; \mb{\phi})$
matches the \emph{variational posterior} $q({\mb{\lambda}}\, | \,\mb{z}; {\mb{\theta}})$. From
this perspective, we can view $r$ as a {recursive variational approximation}.
It is a model for the {posterior} $q$ of the mean-field parameters
${\mb{\lambda}}$ given a realization of the latent variables $\mb... | 1511.02386 | https://arxiv.org/abs/1511.02386 | 2016-06-01 02:01:54 | 1511 | en | 4,096 | 95b757d6-04bb-4ca8-acc7-15250d2db810 | 571 | [
11061,
1124,
3096,
35702,
16827,
5410,
25046,
19914,
279,
1124,
336,
759,
90,
29165,
1663,
44900,
92,
400,
80,
2306,
59,
3096,
35702,
12935,
3417,
59,
11,
760,
1124,
26266,
3096,
90,
89,
11061,
28152,
3096,
35702,
15976,
3417,
8,
12947,... |
\kappa<1$ corresponds to H\"older continuity of the gradient, $\|\nabla f(\boldsymbol{y})-\nabla f(\boldsymbol{x})\|_{\ast}\leq \mu \|\boldsymbol{y}-\boldsymbol{x}\|^{\kappa}$.
\item $\kappa=1$ corresponds to Lipschitz-continuity of the gradient, $\|\nabla f(\boldsymbol{y})-\nabla f(\boldsymbol{x})\|_{\ast}\leq \mu\|\b... | 1811.01903 | https://arxiv.org/abs/1811.01903 | 2019-06-21 02:03:31 | 1811 | en | 4,096 | f17d5731-85fe-4087-b4bb-f11483c50b82 | 7,119 | [
59,
74,
27180,
27,
16,
3,
33210,
311,
472,
2105,
2018,
49006,
315,
279,
20169,
11,
57960,
22957,
77,
370,
4260,
282,
11520,
14824,
18785,
90,
88,
5410,
30529,
77,
370,
4260,
282,
11520,
14824,
18785,
45340,
5410,
59,
91,
15159,
59,
... |
supervised}, \cite{Egarter2014PALDi}, decision tree\cite{Jing2014Non}, graph signal processing\cite{He2018Non}, support vector machines\cite{Lin2010Applying}, and non-negative tensor factorization\cite{Figueiredo2014Electrical}.
More recently, with the current breakthrough of deep neural networks in image classificati... | 1902.02675 | https://arxiv.org/abs/1902.02675 | 2019-03-20 01:15:37 | 1902 | en | 4,096 | 80e7f674-df13-4c4e-a6dd-aee61f099edd | 220 | [
12776,
77990,
2137,
1124,
67075,
90,
36,
70,
5295,
17,
15,
16,
19,
73342,
21685,
2137,
5480,
4916,
59,
67075,
90,
41,
287,
17,
15,
16,
19,
8121,
2137,
4771,
8286,
8692,
59,
67075,
90,
1519,
17,
15,
16,
23,
8121,
2137,
1824,
4621,
... |
{\cal L}(\lambda) & = E_q(\log h(\theta)-\log q_\lambda(\theta)), \label{lbq}
\end{align}
where $E_q(\cdot)$ denotes expectation with respect to $q_\lambda$ and $h(\theta)=p(\theta)p(y|\theta)$. If we differentiate under the integral sign in (\ref{lbq}), the
resulting expression for the gradient can also be written a... | 1801.07873 | https://arxiv.org/abs/1801.07873 | 2018-04-26 02:06:02 | 1801 | en | 4,096 | f2f3c137-1b54-4824-a513-94815bc8c728 | 3,519 | [
28152,
5416,
444,
92,
11520,
12935,
8,
609,
284,
468,
8976,
11520,
839,
305,
11520,
15976,
7287,
59,
839,
2804,
56014,
12935,
11520,
15976,
5731,
1124,
1502,
90,
21123,
80,
532,
59,
408,
90,
6612,
532,
2870,
400,
36,
8976,
11520,
5085... |
) of Theorem \ref{TheoremMain}:
Proposition: There exists a set $\mathcal{E}\subset \mathbb{R}^2$ of Lebesgue measure $0$, so that for all $(x_0,y_0)\in \mathbb{R}^2\backslash \mathcal{E}$, the sequence $(x_{n+1},y_{n+1})=H(x_n,y_n)$ satisfies the following: $(x_n,y_n)$ converges to a critical point of $f$ which is n... | 1911.04221 | https://arxiv.org/abs/1911.04221 | 2019-11-14 02:12:35 | 1911 | en | 4,096 | 6d4394b0-8404-4e6e-adc5-755daebfb22c | 10,775 | [
8,
315,
576,
13173,
1124,
1097,
90,
785,
13173,
6202,
16097,
4710,
2008,
2113,
25,
2619,
6724,
264,
738,
57960,
10374,
5416,
90,
36,
11035,
38253,
1124,
10374,
6066,
90,
49,
92,
61,
17,
3,
315,
1967,
9433,
37668,
6629,
400,
15,
5487... |
the
proposed method outperforms all the previously reported results.
Encouraged by the
good results, we also tested with $N=50$ labels and got a test error
of 1.62~\% ($\pm$ 0.65~\%).
The simple $\Gamma$-model also performed surprisingly well, particularly
for $N=1000$ labels. With $N=100$ labels, all the models som... | 1507.02672 | https://arxiv.org/abs/1507.02672 | 2015-11-25 02:11:02 | 1507 | en | 4,096 | 0331193a-4526-4cf3-adfb-b046fae46488 | 13,254 | [
279,
198,
776,
3865,
1714,
700,
712,
9807,
678,
279,
8597,
4961,
3059,
624,
7408,
413,
3279,
553,
279,
198,
18536,
3059,
11,
582,
1083,
12510,
448,
400,
45,
28,
20,
15,
3,
9201,
323,
2684,
264,
1273,
1465,
198,
1055,
220,
16,
13,
... |
(\rho_{1}=1)$ \\
$\pi_{k} = \text{softmax}(\mathbf{W}_{\mathbf{h}\to\boldsymbol{\pi}}\mathbf{h})_{k}$ \\
$(\Sigma_{0})_{k}=\exp(\mathbf{W}_{\mathbf{h}\to\Sigma_{0}}\mathbf{h})_{k}$ \\
$\Sigma_{k} = (1-\rho_{k}^{2})(\Sigma_{0})_{k} + \tau^{-1}$ \\
$\widetilde{\mathbf{W}}_{k}$ = \cmtt{Cholesky}$(\mathbf{... | 1805.06431 | https://arxiv.org/abs/1805.06431 | 2020-04-09 02:06:18 | 1805 | en | 4,096 | 0d4caeab-7880-4b22-8313-4d39a6976e22 | 6,558 | [
11520,
40322,
15159,
16,
51185,
16,
15087,
90155,
262,
57960,
2493,
15159,
74,
92,
284,
1124,
1318,
90,
77499,
92,
11520,
10374,
13233,
90,
54,
51535,
59,
10374,
13233,
90,
71,
11035,
983,
59,
14824,
18785,
35702,
2493,
3417,
59,
10374,... |
begin{align}
\ensuremath{N}\xspace(\ensuremath{M}\xspace, d) = \Big\{ i \in \ensuremath{\Omega}\xspace \;\Big| \; &\text{FOV}(i, d) \cap \bar{\ensuremath{M}\xspace} \neq \emptyset \text{ and } \\
&\text{FOV}(i, d) \cap \ensuremath{M}\xspace \neq \emptyset\Big\}
\end{align}
Empirically, we find that decreasing $d$ ove... | 2003.00891 | https://arxiv.org/abs/2003.00891 | 2020-03-03 02:31:00 | 2003 | en | 4,096 | 11f0d83b-1cd1-4b25-846a-13e0fd013800 | 107 | [
7265,
90,
6612,
532,
59,
27289,
10374,
90,
45,
11035,
87,
8746,
11520,
27289,
10374,
90,
44,
11035,
87,
8746,
11,
294,
8,
284,
1124,
15636,
59,
90,
600,
1124,
258,
1124,
27289,
10374,
35702,
77335,
11035,
87,
8746,
1124,
17882,
15636,... |
)}
\end{equation}
is the most powerful test statistic.
In order to evaluate $\lambda({\cal D})$, one must be able to evaluate the probability
densities $p_\mathbf{X}(\mathbf{x}| \theta_0)$ and $p_\mathbf{X}(\mathbf{x}| \theta_1)$ at any value $\mathbf{x}$. However,
it is increasingly common in science that one has a c... | 1506.02169 | https://arxiv.org/abs/1506.02169 | 2016-03-21 01:11:07 | 1506 | en | 4,096 | 02c9827e-0c36-460e-8ba2-91f10397dfaf | 1,121 | [
10569,
59,
408,
90,
25310,
367,
532,
285,
279,
1429,
7988,
1273,
42489,
382,
641,
1973,
311,
15442,
57960,
12935,
2306,
59,
5416,
422,
5410,
54876,
825,
1969,
387,
2952,
311,
15442,
279,
18927,
198,
52221,
1361,
400,
79,
56014,
10374,
... |
Subsets and Splits
No community queries yet
The top public SQL queries from the community will appear here once available.