-
Notifications
You must be signed in to change notification settings - Fork 0
/
Copy pathindex.html
787 lines (714 loc) · 41 KB
/
index.html
1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16
17
18
19
20
21
22
23
24
25
26
27
28
29
30
31
32
33
34
35
36
37
38
39
40
41
42
43
44
45
46
47
48
49
50
51
52
53
54
55
56
57
58
59
60
61
62
63
64
65
66
67
68
69
70
71
72
73
74
75
76
77
78
79
80
81
82
83
84
85
86
87
88
89
90
91
92
93
94
95
96
97
98
99
100
101
102
103
104
105
106
107
108
109
110
111
112
113
114
115
116
117
118
119
120
121
122
123
124
125
126
127
128
129
130
131
132
133
134
135
136
137
138
139
140
141
142
143
144
145
146
147
148
149
150
151
152
153
154
155
156
157
158
159
160
161
162
163
164
165
166
167
168
169
170
171
172
173
174
175
176
177
178
179
180
181
182
183
184
185
186
187
188
189
190
191
192
193
194
195
196
197
198
199
200
201
202
203
204
205
206
207
208
209
210
211
212
213
214
215
216
217
218
219
220
221
222
223
224
225
226
227
228
229
230
231
232
233
234
235
236
237
238
239
240
241
242
243
244
245
246
247
248
249
250
251
252
253
254
255
256
257
258
259
260
261
262
263
264
265
266
267
268
269
270
271
272
273
274
275
276
277
278
279
280
281
282
283
284
285
286
287
288
289
290
291
292
293
294
295
296
297
298
299
300
301
302
303
304
305
306
307
308
309
310
311
312
313
314
315
316
317
318
319
320
321
322
323
324
325
326
327
328
329
330
331
332
333
334
335
336
337
338
339
340
341
342
343
344
345
346
347
348
349
350
351
352
353
354
355
356
357
358
359
360
361
362
363
364
365
366
367
368
369
370
371
372
373
374
375
376
377
378
379
380
381
382
383
384
385
386
387
388
389
390
391
392
393
394
395
396
397
398
399
400
401
402
403
404
405
406
407
408
409
410
411
412
413
414
415
416
417
418
419
420
421
422
423
424
425
426
427
428
429
430
431
432
433
434
435
436
437
438
439
440
441
442
443
444
445
446
447
448
449
450
451
452
453
454
455
456
457
458
459
460
461
462
463
464
465
466
467
468
469
470
471
472
473
474
475
476
477
478
479
480
481
482
483
484
485
486
487
488
489
490
491
492
493
494
495
496
497
498
499
500
501
502
503
504
505
506
507
508
509
510
511
512
513
514
515
516
517
518
519
520
521
522
523
524
525
526
527
528
529
530
531
532
533
534
535
536
537
538
539
540
541
542
543
544
545
546
547
548
549
550
551
552
553
554
555
556
557
558
559
560
561
562
563
564
565
566
567
568
569
570
571
572
573
574
575
576
577
578
579
580
581
582
583
584
585
586
587
588
589
590
591
592
593
594
595
596
597
598
599
600
601
602
603
604
605
606
607
608
609
610
611
612
613
614
615
616
617
618
619
620
621
622
623
624
625
626
627
628
629
630
631
632
633
634
635
636
637
638
639
640
641
642
643
644
645
646
647
648
649
650
651
652
653
654
655
656
657
658
659
660
661
662
663
664
665
666
667
668
669
670
671
672
673
674
675
676
677
678
679
680
681
682
683
684
685
686
687
688
689
690
691
692
693
694
695
696
697
698
699
700
701
702
703
704
705
706
707
708
709
710
711
712
713
714
715
716
717
718
719
720
721
722
723
724
725
726
727
728
729
730
731
732
733
734
735
736
737
738
739
740
741
742
743
744
745
746
747
748
749
750
751
752
753
754
755
756
757
758
759
760
761
762
763
764
765
766
767
768
769
770
771
772
773
774
775
776
777
778
779
780
781
782
783
784
785
786
787
<!DOCTYPE HTML>
<!--
Hyperspace by HTML5 UP
html5up.net | @ajlkn
Free for personal and commercial use under the CCA 3.0 license (html5up.net/license)
-->
<html>
<script src='https://cdnjs.cloudflare.com/ajax/libs/mathjax/2.7.2/MathJax.js?config=TeX-MML-AM_CHTML'></script>
<head>
<!-- <script type=“text/javascript” async src=“https://cdnjs.cloudflare.com/ajax/libs/mathjax/2.7.2/MathJax.js?config=TeX-MML-AM_CHTML“> -->
<!-- <script type=“text/javascript” async src=“https://unpkg.com/ajax/libs/mathjax/2.7.2/MathJax.js?config=TeX-MML-AM_CHTML“>
</script> -->
<!-- // to enable MathJax writing -->
<title>(Almost) 50 shades of Bayesian Learning: PAC-Bayesian trends and insights - NIPS 2017 workshop</title>
<meta charset="utf-8" />
<meta name="viewport" content="width=device-width, initial-scale=1" />
<!--[if lte IE 8]><script src="assets/js/ie/html5shiv.js"></script><![endif]-->
<link rel="stylesheet" href="assets/css/main.css" />
<!--[if lte IE 9]><link rel="stylesheet" href="assets/css/ie9.css" /><![endif]-->
<!--[if lte IE 8]><link rel="stylesheet" href="assets/css/ie8.css" /><![endif]-->
</head>
<body>
<!-- Sidebar -->
<section id="sidebar">
<div class="inner">
<nav>
<ul>
<li><a href="#intro">Home</a></li>
<li><a href="#one">Scope</a></li>
<li><a href="#two">Speakers</a></li>
<li><a href="#three">Schedule</a></li>
<!-- <li><a href="#three">Accepted papers</a></li> -->
<li><a href="#four">CFP & Dates</a></li>
<li><a href="#six">Accepted Papers</a></li>
<li><a href="#seven">Material: slides, videos and pictures</a></li>
<li><a href="#five">Organizers</a></li>
</ul>
</nav>
</div>
</section>
<!-- Wrapper -->
<div id="wrapper">
<!-- Intro -->
<section id="intro" class="wrapper style1 fullscreen fade-up">
<div class="inner">
<h1><a href="https://nips.cc">NIPS 2017</a> Workshop</h1>
<h2>(Almost) 50 Shades of Bayesian Learning: PAC-Bayesian trends and insights</h2> <br> <a href="https://www.google.com/maps/place/Long+Beach+Convention+%26+Entertainment+Center/@33.7606839,-118.1892951,16z/data=!3m1!5s0x80dd313b1d738beb:0xb11de026a4091d6e!4m2!3m1!1s0x80dd313b68c4eae7:0x69f1fff3cb508d42">Long Beach Convention Center, California</a> <br> <b>Room 101-A</b> <br>December 9, 2017 <br> <a href="#three"><b>Schedule</b></a> <br><a href="https://www.youtube.com/watch?v=AmdKNkBOveM&list=PLBLqzR5n5bIioJRyYDUqtCW0YfWpWoZg1">YouTube channel</a>
<!-- <ul class="actions">
<li><a href="#one" class="button scrolly">Learn more</a></li>
</ul> -->
</div>
</section>
<!-- <section id="intro2" class="wrapper style2 spotlights">
</section> -->
<!-- One -->
<section id="one" class="wrapper style2 spotlights">
<section>
<!-- <a href="" class="image"><img src="images/pic01.jpg" alt="" data-position="center center" /></a> -->
<div class="content">
<div class="inner">
<h2>Scope</h2>
<p align="justify">Industry-wide successes of machine learning at the dawn of the (so-called) big data era has led to an increasing gap between practitioners and theoreticians. The former are using off-the-shelf statistical and machine learning methods, while the latter are designing and studying the mathematical properties of such algorithms. The tradeoff between those two movements is somewhat addressed by Bayesian researchers, where sound mathematical guarantees often meet efficient implementation and provide model selection criteria. In the late 90s, a new paradigm has emerged in the statistical learning community, used to derive probably approximately correct (PAC) bounds on Bayesian-flavored estimators. This PAC-Bayesian theory has been pioneered by Shawe-Taylor and Willamson (1997), and McAllester (1998, 1999). It has been extensively formalized by Catoni (2004, 2007) and has triggered, slowly but surely, increasing research efforts during last decades.</p>
<p align="justify">We believe it is time to pinpoint the current PAC-Bayesian trends relatively to other modern approaches in the (statistical) machine learning community. Indeed, we observe that, while the field grows by its own, it took some undesirable distance from some related areas. Firstly, it seems to us that the relation to Bayesian methods has been forsaken in numerous works, despite the potential of PAC-Bayesian theory to bring new insights to the Bayesian community and to go beyond the classical Bayesian/frequentist divide. Secondly, the PAC-Bayesian methods share similarities with other quasi-Bayesian (or pseudo-Bayesian) methods studying Bayesian practices from a frequentist standpoint, such as the Minimum Description Length (MDL) principle (Grünwald, 2007). Last but not least, even if some practical and theory grounded learning algorithm has emerged from PAC-Bayesian works, these are almost unused for real-world problems.</p>
<p align="justify">In short, this workshop aims at gathering statisticians and machine learning researchers to discuss current trends and the future of {PAC,quasi}-Bayesian learning. From a broader perspective, we aim to bridge the gap between several communities that can all benefit from sharper statistical guarantees and sound theory-driven learning algorithms.</p>
<h3>References</h3>
<ol>
<li>
J. Shawe-Taylor and R. Williamson. <a href="http://dl.acm.org/citation.cfm?id=267466">A PAC analysis of a Bayes estimator</a>. In Proceedings of COLT, 1997.
</li>
<li>
D. A. McAllester. <a href="http://dl.acm.org/citation.cfm?id=279989">Some PAC-Bayesian theorems</a>. In Proceedings of COLT, 1998.
</li>
<li>
D. A. McAllester. <a href="http://dl.acm.org/citation.cfm?id=307435">PAC-Bayesian model averaging</a>. In Proceedings of COLT, 1999.
</li>
<li>
O. Catoni. <a href="http://www.math.ens.fr/~catoni/homepage/saintFlourDraft.ps">Statistical Learning Theory and Stochastic Optimization</a>. Saint-Flour Summer School on Probability Theory 2001 (Jean Picard ed.), Lecture Notes in Mathematics. Springer, 2004.
</li>
<li>
O. Catoni. <a href="https://arxiv.org/abs/0712.0248">PAC-Bayesian supervised classification: the thermodynamics of statistical learning</a>. Institute of Mathematical Statistics Lecture Notes—Monograph Series, 56. Institute of Mathematical Statistics, 2007.
</li>
<li>
P. D. Grünwald. <a href="https://www.cwi.nl/~pdg/book/book.html">The Minimum Description Length Principle</a>. The MIT Press, 2007.
</li>
</ol>
<!-- <ul class="actions">
<li><a href="#" class="button">Learn more</a></li>
</ul> -->
</div>
</div>
</section>
<!-- <section>
<a href="#" class="image"><img src="images/pic01.jpg" alt="" data-position="center center" /></a>
<div class="content">
<div class="inner">
<h2>Sed ipsum dolor</h2>
<ul class="actions">
<li><a href="#" class="button">Learn more</a></li>
</ul>
</div>
</div>
</section> -->
<!-- <section>
<a href="#" class="image"><img src="images/pic01.jpg" alt="" data-position="center center" /></a>
<div class="content">
<div class="inner">
<h2>Sed ipsum dolor</h2>
<ul class="actions">
<li><a href="#" class="button">Learn more</a></li>
</ul>
</div>
</div>
</section> -->
<!--
<section>
<a href="#" class="image"><img src="images/pic02.jpg" alt="" data-position="top center" /></a>
<div class="content">
<div class="inner">
<h2>References</h2>
<ul class="actions">
<li><a href="#" class="button">Learn more</a></li>
</ul>
</div>
</div>
</section> -->
<!-- <section>
<a href="#" class="image"><img src="images/pic03.jpg" alt="" data-position="25% 25%" /></a>
<div class="content">
<div class="inner">
<h2>Ultricies aliquam</h2>
<p>Phasellus convallis elit id ullamcorper pulvinar. Duis aliquam turpis mauris, eu ultricies erat malesuada quis. Aliquam dapibus.</p>
<ul class="actions">
<li><a href="#" class="button">Learn more</a></li>
</ul>
</div>
</div>
</section> -->
</section>
<!-- Two -->
<section id="two" class="wrapper style3 fade-up">
<div class="inner">
<h2>Invited Speakers</h2>
<p>The workshop welcomes world-class experts on {quasi,PAC,∅}-Bayesian learning.</p>
<div class="features">
<section align="center">
<!-- <span class="icon major fa-code"></span> -->
<span class="image"><img src="images/catoni.jpg" height="128" width="128" alt="" data-position="center center" /></span>
<h3><a href="http://ocatoni.perso.math.cnrs.fr/">Olivier Catoni</a></h3>
<p>Senior researcher, CNRS, France.</p>
</section>
<section align="center">
<span class="image"><img src="images/grunwald.jpg" height="128" width="128" alt="" data-position="center center" /></span>
<h3><a href="http://homepages.cwi.nl/~pdg/">Peter Grünwald</a></h3>
<p>Professor, CWI, The Netherlands.</p>
</section>
<section align="center">
<span class="image"><img src="images/laviolette.jpg" height="128" width="128" alt="" data-position="center center" /></span>
<h3><a href="http://www2.ift.ulaval.ca/~laviolette/">François Laviolette</a></h3>
<p>Professor, Université Laval, Canada.</p>
</section>
<section align="center">
<span class="image"><img src="images/lawrence.jpg" height="128" width="128" alt="" data-position="center center" /></span>
<h3><a href="http://inverseprobability.com">Neil Lawrence</a></h3>
<p>Professor, University of Sheffield, and<br>Amazon Research Cambridge, UK.</p>
</section>
<section align="center">
<span class="image"><img src="images/marin.jpg" height="128" width="128" alt="" data-position="center center" /></span>
<h3><a href="http://www.math.univ-montp2.fr/~marin/">Jean-Michel Marin</a></h3>
<p>Professor, Université de Montpellier, France.</p>
</section>
<section align="center">
<span class="image"><img src="images/roy.jpg" height="128" width="128" alt="" data-position="center center" /></span>
<h3><a href="http://danroy.org">Dan Roy</a></h3>
<p>Assistant Professor, University of Toronto, Canada.</p>
</section>
<section align="center">
<span class="image"><img src="images/seldin.jpg" height="128" width="128" alt="" data-position="center center" /></span>
<h3><a href="https://sites.google.com/site/yevgenyseldin/">Yevgeny Seldin</a></h3>
<p>Associate Professor, University of Copenhagen, Denmark.</p>
</section>
<section align="center">
<span class="image"><img src="images/shawe-taylor.jpg" height="128" width="128" alt="" data-position="center center" /></span>
<h3><a href="http://www0.cs.ucl.ac.uk/staff/J.Shawe-Taylor/">John Shawe-Taylor</a></h3>
<p>Professor, University College London, UK.</p>
</section>
<section align="center">
<span class="image"><img src="images/teh.jpg" height="128" width="128" alt="" data-position="center center" /></span>
<h3><a href="https://www.stats.ox.ac.uk/~teh/">Yee-Whye Teh</a></h3>
<p>Professor, University of Oxford, UK.</p>
</section>
</div>
<!-- <ul class="actions">
<li><a href="#" class="button">Learn more</a></li>
</ul> -->
</div>
</section>
<!-- Three -->
<section id="three" class="wrapper style1 fade-up">
<div class="inner">
<h2>Schedule</h2>
The workshop will run in <b>room 101-A</b>.
<p><a href="https://nips.cc/Conferences/2017/Schedule?showEvent=8760">Check NIPS website.</a></p>
<table style="width:100%">
<tr>
<th>Time</th>
<th>Title</th>
<th>Speaker</th>
</tr>
<tr>
<td>8.30</td>
<td>Overture</td>
<td>Organizers</td>
</tr>
<tr>
<td>8.30-9.30</td>
<td>
<!-- <details> -->
<!-- <summary> -->
A Tutorial on PAC-Bayesian Theory
<!-- </summary> -->
<!-- <p>
A tutorial on PAC-Bayesian Theory.
</p> -->
<!-- </details> -->
</td>
<td><a href="http://www2.ift.ulaval.ca/~laviolette/">François Laviolette</a></td>
</tr>
<tr>
<td>9.30-10.15</td>
<td>
<details>
<summary>
A Tight Excess Risk Bound via a Unified PAC-Bayesian-Rademacher-Shtarkov-MDL Complexity
</summary>
<p>
Over the last 15 years, machine learning theorists have bounded the performance of empirical risk minimization by (localized) Rademacher complexity; Bayesians with frequentist sympathies have studied Bayesian consistency and rate of convergence theorems, sometimes under misspecification, and PAC-Bayesians have studied convergence properties of generalized Bayesian and Gibbs posteriors. We show that, amazingly, most such bounds readily follow from essentially a single result that bounds excess risk in terms of a novel complexity COMP\( (\eta,w) \). which depends on a learning rate \( \eta \) and a luckiness function \( w \), the latter generalizing the concept of a 'prior'. Depending on the choice of \( w \), COMP\( (\eta,w) \) specializes to PAC-Bayesian (KL(posterior||prior) complexity, MDL (normalized maximum likelihood) complexity and Rademacher complexity, and the bounds obtained are optimized for generalized Bayes, ERM, penalized ERM (such as Lasso) or other methods. Tuning \( \eta \) leads to optimal excess risk convergence rates, even for very large (polynomial entropy) classes which have always been problematic for the PAC-Bayesian approach; the optimal \( \eta \) depends on 'fast rate' properties of the domain, such as central, Bernstein and Tsybakov conditions.
<br><br>
Joint work with Nishant Mehta, University of Victoria. See <a href="https://arxiv.org/abs/1710.07732">the paper</a>.
</p>
</details>
</td>
<td><a href="http://homepages.cwi.nl/~pdg/">Peter Grünwald</a></td>
</tr>
<tr>
<td>10.15-11.00</td>
<td><b>Coffee break and posters session</b></td>
<td></td>
</tr>
<tr>
<td>11.00-11.45</td>
<td>
<details>
<summary>
Some recent advances on Approximate Bayesian Computation techniques
</summary>
<p>
In an increasing number of application domains, the statistical model is so complex that the point-wise computation of the likelihood is intractable. That is typically the case when the underlying probability distribution involves numerous latent variables. Approximate Bayesian Computation (ABC) is a widely used technique to bypass that difficulty. I will review some recent developments on ABC techniques, emphazing the fact that modern machine learning approaches are useful in this field. Although intrinsically very different of PAC-Bayesian strategies - the choice of a generative model is essential within the ABC paradigm - I will highlight some links between these two methodologies.
</p>
</details>
</td>
<td><a href="http://www.math.univ-montp2.fr/~marin/">Jean-Michel Marin</a></td>
</tr>
<tr>
<td>11.45-12.05</td>
<td>
<details>
<summary>
Contributed talk: A PAC-Bayesian Approach to Spectrally-Normalized Margin Bounds for Neural Networks
</summary>
<p>
We present a generalization bound for feedforward neural networks in terms of the product of the spectral norm of the layers and the Frobenius norm of the weights. The generalization bound is derived using a PAC-Bayes analysis.
</p>
</details>
</td>
<td><a href="http://www.math.ias.edu/~bneyshabur/">Behnam Neyshabur</a></td>
</tr>
<tr>
<td>12.05-14.00</td>
<td><b>Lunch break</b></td>
<td></td>
</tr>
<tr>
<td>14.00-14.40</td>
<td>
<details>
<summary>
Dimension-free PAC-Bayesian Bounds
</summary>
<p>
PAC-Bayesian inequalities have already proved to be a great tool
to obtain dimension free generalization bounds, such as margin
bounds for Support Vector Machines. In this talk,
we will play with PAC-Bayesian inequalities and influence
functions to present new robust estimators for the mean
of random vectors and random matrices, as well as for
linear least squares regression. A common theme of the
presentation will be to establish dimension free bounds
and to work under mild polynomial moment assumptions
regarding the tail of the sample distribution.
</p>
</details>
</td>
<td><a href="http://ocatoni.perso.math.cnrs.fr/">Olivier Catoni</a></td>
</tr>
<tr>
<td>14.40-15.00</td>
<td>
<details>
<summary>
Contributed talk: Dimension free PAC-Bayesian bounds for the estimation of the mean of a random vector
</summary>
<p>
We present a new estimator of the mean of a random vector, computed by applying some threshold function to the norm. Non asymptotic dimension-free almost sub-Gaussian bounds are proved under weak moment assumptions, using PAC-Bayesian inequalities.
</p>
</details>
</td>
<td><a href="http://ocatoni.perso.math.cnrs.fr/">Olivier Catoni</a></td>
</tr>
<tr>
<td>15.00-15.30</td>
<td><b>Coffee break and posters session</b></td>
<td></td>
</tr>
<tr>
<td>15.30-16.15</td>
<td>
<details>
<summary>
A Strongly Quasiconvex PAC-Bayesian Bound
</summary>
<p>
We propose a new PAC-Bayesian bound and a way of constructing a hypothesis space, so that the bound is convex in the posterior distribution and also convex in a trade-off parameter between empirical performance of the posterior distribution and its complexity. The complexity is measured by the Kullback-Leibler divergence to a prior. We derive an alternating procedure for minimizing the bound. We show that the bound can be rewritten as a one-dimensional function of the trade-off parameter and provide sufficient conditions under which the function has a single global minimum. When the conditions are satisfied the alternating minimization is guaranteed to converge to the global minimum of the bound. We provide experimental results demonstrating that rigorous minimization of the bound is competitive with cross-validation in tuning the trade-off between complexity and empirical performance. In all our experiments the trade-off turned to be quasiconvex even when the sufficient conditions were violated.
<br><br>
Joint work with Niklas Thiemann, Christian Igel, and Olivier Wintenberger.
</p>
</details>
</td>
<td><a href="https://sites.google.com/site/yevgenyseldin/">Yevgeny Seldin</a></td>
</tr>
<tr>
<td>16.15-17.00</td>
<td>
<details>
<summary>
Distribution Dependent Priors for Stable Learning
</summary>
<p>
One feature of PAC-Bayes approach is that though the prior must be fixed, we do not need to have an explicit expression for it, only to be able to bound the distance between prior and posterior. Furthermore, the choice of prior only impacts the quality of the bound and not the validity of the results. We will discuss the implications of these observations describing ways in which the prior may be chosen to improve the quality of the bounds obtained. The application of these ideas to the stability analysis for SVMs delivers a tightening of the well-known stability bounds.
</p>
</details>
</td>
<td><a href="http://www0.cs.ucl.ac.uk/staff/J.Shawe-Taylor/">John Shawe-Taylor</a></td>
</tr>
<tr>
<td>17.00-17.30</td>
<td>
<details>
<summary>
Deep Neural Networks: From Flat Minima to Numerically Nonvacuous Generalization Bounds via PAC-Bayes
</summary>
<p>
One of the defining properties of deep learning is that models are chosen to have many more parameters than available training data. In light of this capacity for overfitting, it is remarkable that simple algorithms like SGD reliably return solutions with low test error. One roadblock to explaining these phenomena in terms of implicit regularization, structural properties of the solution, and/or easiness of the data is that many learning bounds are quantitatively vacuous when applied to networks learned by SGD in this "deep learning" regime. Logically, in order to explain generalization, we need nonvacuous bounds.
<br><br>
I will discuss recent work using PAC-Bayesian bounds and optimization to arrive at nonvacuous generalization bounds for neural networks with millions of parameters trained on only tens of thousands of examples. We connect our findings to recent and old work on flat minima and MDL-based explanations of generalization, as well as to variational inference for deep learning. Time permitting, I'll discuss new work interpreting Entropy-SGD as a PAC-Bayesian method.
<br><br>
Joint work with Gintare Karolina Dziugaite, based on <a href="https://arxiv.org/abs/1703.11008">this paper</a>.
</p>
</details>
</td>
<td><a href="http://danroy.org">Dan Roy</a></td>
</tr>
<tr>
<td>17.30-18.25</td>
<td>Discussion (moderated by Francis Bach)</td>
<td>Panelists: Olivier Catoni, Peter Grünwald, François Laviolette, Jean-Michel Marin, Yevgeny Seldin, John Shawe-Taylor, Yee-Whye Teh</td>
</tr>
<tr>
<td>18.25</td>
<td>Concluding remarks</td>
<td>Organizers</td>
</tr>
</table>
<!-- <div class="split style1">
<section>
<form method="post" action="#">
<div class="field half first">
<label for="name">Name</label>
<input type="text" name="name" id="name" />
</div>
<div class="field half">
<label for="email">Email</label>
<input type="text" name="email" id="email" />
</div>
<div class="field">
<label for="message">Message</label>
<textarea name="message" id="message" rows="5"></textarea>
</div>
<ul class="actions">
<li><a href="" class="button submit">Send Message</a></li>
</ul>
</form>
</section>
<section>
<ul class="contact">
<li>
<h3>Address</h3>
<span>12345 Somewhere Road #654<br />
Nashville, TN 00000-0000<br />
USA</span>
</li>
<li>
<h3>Email</h3>
<a href="#">user@untitled.tld</a>
</li>
<li>
<h3>Phone</h3>
<span>(000) 000-0000</span>
</li>
<li>
<h3>Social</h3>
<ul class="icons">
<li><a href="#" class="fa-twitter"><span class="label">Twitter</span></a></li>
<li><a href="#" class="fa-facebook"><span class="label">Facebook</span></a></li>
<li><a href="#" class="fa-github"><span class="label">GitHub</span></a></li>
<li><a href="#" class="fa-instagram"><span class="label">Instagram</span></a></li>
<li><a href="#" class="fa-linkedin"><span class="label">LinkedIn</span></a></li>
</ul>
</li>
</ul>
</section>
</div> -->
</div>
</section>
<section id="four" class="wrapper style3 fade-up">
<div class="inner">
<h2>Call for Papers</h2>
<p>All accepted papers will have a poster presentation, and we will select two papers for oral presentations. We welcome submissions related to the following list of topics:
<ul>
<li>
{PAC,quasi,∅}-Bayesian generalization guarantees,
</li>
<li>
Novel theoretical perspectives on Bayesian methods,
</li>
<li>
Application of the PAC-Bayesian theory to different learning frameworks,
</li>
<li>
Learning algorithms inspired by a {PAC,quasi}-Bayesian analysis.
</li>
</ul>
<h3>
Submission Instructions
</h3>
<p>
Submission is no longer possible. </p>
<!-- <a href="https://easychair.org/conferences/?conf=pacbayes2017">Click here to submit a paper</a>. -->
<ul>
<li>
Page limit: 4 pages (excluding references).
</li>
<li>
Please use <a href="https://nips.cc/Conferences/2017/PaperInformation/StyleFiles">the NIPS 2017 submission format</a>.
</li>
<li>
We are committed to a double-blind reviewing process. Submissions must be made anonymous: please hide authors' names and affiliations.
</li>
<li>
Results which were previously published may be submitted, although we encourage original contributions. Please clearly indicate if the submitted work has been presented or published elsewhere.
</li>
<li>
Please note that at least one author of each accepted paper must be available to present the paper at the workshop.
</li>
</ul>
<!-- <h3>Important Dates</h3>
<table align="center">
<tr align="center">
<th>
Deadline for submission of papers
</th>
<th>
Notification of acceptance
</th>
</tr>
<tr align="center">
<td>
October 27, 2017, 23:59 PDT (UTC-8)
<iframe src="http://free.timeanddate.com/countdown/i5w1g8ar/n137/cf12/cm0/cu4/ct0/cs1/ca0/co0/cr0/ss0/cac000/cpcf00/pct/tcfff/fs100/szw320/szh135/iso2017-10-08T00:00:00" allowTransparency="true" frameborder="0" width="113" height="30"></iframe>
</td>
<td>
November 9, 2017
</td>
</tr>
</table> -->
<h3>Scientific Committee</h3>
<ul>
<li>
<a href="http://alquier.ensae.net">Pierre Alquier</a>, Université Paris-Saclay, France
</li>
<li>
<a href="http://www.lpma-paris.fr/pageperso/castillo/">Ismaël Castillo</a>, Université Pierre et Marie Curie, France
</li><li>
<a href="http://perso.univ-st-etienne.fr/habrarda/">Amaury Habrard</a>, Université Jean Monnet, France
</li><li>
<a href="http://www.cs.umd.edu/~blondon/">Ben London</a>, Amazon, USA
</li><li>
<a href="http://perso.univ-st-etienne.fr/me63854h/">Emilie Morvant</a>, Université Jean Monnet, France
</li><li>
<a href="http://www.is.titech.ac.jp/~s-taiji/">Taiji Suzuki</a>, Tokyo Institute of Technology, Japan
</li>
</ul>
</section>
<section id="six" class="wrapper style2 fade-up">
<div class="inner">
<h2>Accepted papers</h2>
<!-- <p>All accepted papers will have a poster presentation, and we will select two papers for oral presentations. We welcome submissions related to the following list of topics: -->
<h3>
Contributed talks
</h3>
<ol>
<li>
<a href="pdf/PAC-Bayes_2017_paper_2.pdf"><em>A PAC-Bayesian Approach to Spectrally-Normalized Margin Bounds for Neural Networks</em></a> <br> Behnam Neyshabur, Srinadh Bhojanapalli, David McAllester and Nathan Srebro
</li>
<li>
<a href="pdf/PAC-Bayes_2017_paper_1.pdf"><em>Dimension free PAC-Bayesian bounds for the estimation of the mean of a random vector</em></a>
<br><a href="http://ocatoni.perso.math.cnrs.fr">Olivier Catoni</a> and <a href="http://pages.saclay.inria.fr/ilaria.giulini/">Ilaria Giulini</a>
</li>
</ol>
<h3>
Posters
</h3>
<ol>
<li>
<a href="http://papers.nips.cc/paper/6853-multi-way-interacting-regression-via-factorization-machines"><em>Multi-way Interacting Regression via Factorization Machines</em></a> <br> Mikhail Yurochkin, XuanLong Nguyen and Nikolaos Vasiloglou
</li>
<li>
<a href="http://papers.nips.cc/paper/6785-lookahead-bayesian-optimization-with-inequality-constraints"><em>Lookahead Bayesian Optimization with Inequality Constraints</em></a> <br> Remi Lam and Karen Willcox
</li>
<li>
<a href="http://papers.nips.cc/paper/7059-convergence-rates-of-a-partition-based-bayesian-multivariate-density-estimation-method"><em>Convergence rates of a partition based Bayesian multivariate density estimation method</em></a> <br> Linxi Liu, Dangna Li and Wing Hung Wong
</li>
<li>
<a href="http://papers.nips.cc/paper/7111-bayesian-optimization-with-gradients"><em>Bayesian Optimization with Gradients</em></a> <br> Jian Wu, Matthias Poloczek, Andrew Gordon Wilson and Peter I. Frazier
</li>
<li>
<a href="http://papers.nips.cc/paper/7219-simple-and-scalable-predictive-uncertainty-estimation-using-deep-ensembles"><em>Simple and Scalable Predictive Uncertainty Estimation using Deep Ensembles</em></a> <br> Balaji Lakshminarayanan, Alexander Pritzel and Charles Blundell
</li>
<li>
<a href="http://papers.nips.cc/paper/6725-adaptive-bayesian-sampling-with-monte-carlo-em"><em>Adaptive Bayesian Sampling with Monte Carlo EM</em></a> <br> Anirban Roychowdhury and Srinivasan Parthasarathy
</li>
<li>
<a href="http://papers.nips.cc/paper/7100-excess-risk-bounds-for-the-bayes-risk-using-variational-inference-in-latent-gaussian-models"><em>Excess Risk Bounds for the Bayes Risk using Variational Inference in Latent Gaussian Models</em></a> <br> Rishit Sheth and Roni Khardon
</li>
<li>
<a href="http://papers.nips.cc/paper/7176-exploring-generalization-in-deep-learning"><em>Exploring Generalization in Deep Learning</em></a> <br> Behnam Neyshabur, Srinadh Bhojanapalli, David McAllester and Nathan Srebro
</li>
<li>
<a href="http://papers.nips.cc/paper/6886-a-pac-bayesian-analysis-of-randomized-learning-with-application-to-stochastic-gradient-descent"><em>A PAC-Bayesian Analysis of Randomized Learning with Application to Stochastic Gradient Descent</em></a> <br> Ben London
</li>
<li>
<a href="http://papers.nips.cc/paper/7254-structured-bayesian-pruning-via-log-normal-multiplicative-noise"><em>Structured Bayesian Pruning via Log-Normal Multiplicative Noise</em></a> <br>Kirill Neklyudov, Dmitry Molchanov, Arsenii Ashukha and Dmitry Vetrov
</li>
<li>
<a href="pdf/PAC-Bayes_2017_paper_3.pdf"><em>PAC-Bayesian Generalization Bound for Multi-class Learning</em></a> <br> Loubna Benabbou and Pascal Lang
</li>
</ol>
<!-- <h3>Important Dates</h3>
<table align="center">
<tr align="center">
<th>
Deadline for submission of papers
</th>
<th>
Notification of acceptance
</th>
</tr>
<tr align="center">
<td>
October 27, 2017, 23:59 PDT (UTC-8)
<iframe src="http://free.timeanddate.com/countdown/i5w1g8ar/n137/cf12/cm0/cu4/ct0/cs1/ca0/co0/cr0/ss0/cac000/cpcf00/pct/tcfff/fs100/szw320/szh135/iso2017-10-08T00:00:00" allowTransparency="true" frameborder="0" width="113" height="30"></iframe>
</td>
<td>
November 9, 2017
</td>
</tr>
</table> -->
</section>
<section id="seven" class="wrapper style3 fade-up">
<div class="inner">
<h2>Material</h2>
<!-- <p>All accepted papers will have a poster presentation, and we will select two papers for oral presentations. We welcome submissions related to the following list of topics: -->
<p>
Slides are available below. A YouTube channel hosts most of the videos of the workshop: <a href="https://www.youtube.com/watch?v=AmdKNkBOveM&list=PLBLqzR5n5bIioJRyYDUqtCW0YfWpWoZg1">go to YouTube</a>.
</p>
<!-- <p>
Benjamin Guedj's YouTube channel: <a href="https://www.youtube.com/channel/UCkp0b5jlZoIc4U-sDav4Fuw"></a>
</p> -->
<ul>
<li>
Olivier Catoni - Dimension-free PAC-Bayesian Bounds <br>
[ <a href="pdf/catoni_nips2017_1.pdf">Slides 1/2</a> ] • [ <a href="pdf/catoni_nips2017_2.pdf">Slides 2/2</a> ] • [ <a href="https://youtu.be/KtRP4yPd-9Q">Video</a> ] • [ <a href="images/live_catoni.jpg">Picture</a> ]
</li>
<li>
Peter Grünwald - A Tight Excess Risk Bound via a Unified PAC-Bayesian-Rademacher-Shtarkov-MDL Complexity <br>
[ <a href="pdf/grunwald_nips2017.pdf">Slides</a> ] • [ <a href="https://youtu.be/O2qpGBw8Y6w">Video</a> ] • [ <a href="images/live_grunwald.jpg">Picture</a> ]
</li>
<li>
François Laviolette - A Tutorial on PAC-Bayesian Theory <br>
[ <a href="pdf/laviolette_nips2017.pdf">Slides</a> ] • [ <a href="https://youtu.be/GnRX9Pvw6Xw">Video</a> ] • [ <a href="images/live_laviolette.jpg">Picture</a> ]
</li>
<li>
Jean-Michel Marin - Some recent advances on Approximate Bayesian Computation techniques <br>
[ <a href="pdf/marin_nips2017.pdf">Slides</a> ] • [ <a href="https://youtu.be/AmdKNkBOveM">Video</a> ] • [ <a href="images/live_marin.jpg">Picture</a> ]
</li>
<li>
Behnam Neyshabur - A PAC-Bayesian Approach to Spectrally-Normalized Margin Bounds for Neural Networks <br>
[ <a href="pdf/neyshabur_nips2017.pdf">Slides</a> ] • [ <a href="https://youtu.be/I7qfkZEQsIA">Video</a> ] • [ <a href="images/live_neyshabur.jpg">Picture</a> ]
</li>
<li>
Dan Roy - Deep Neural Networks: From Flat Minima to Numerically Nonvacuous Generalization Bounds via PAC-Bayes <br>
[ <a href="pdf/roy_nips2017.pdf">Slides</a> ] • [ <a href="https://youtu.be/fAUf8F5mLAQ">Video</a> ] • [ <a href="images/live_roy.jpg">Picture</a> ]
</li>
<li>
Yevgeny Seldin - A Strongly Quasiconvex PAC-Bayesian Bound <br>
[ <a href="pdf/seldin_nips2017.pdf">Slides</a> ] • [ <a href="https://youtu.be/DN34OW8rz1Y">Video</a> ] • [ <a href="images/live_seldin.jpg">Picture</a> ]
</li>
<li>
John Shawe-Taylor - Distribution Dependent Priors for Stable Learning <br>
[ <a href="pdf/shawe-taylor_nips2017.pdf">Slides</a> ] • [ <a href="https://youtu.be/1WlFPR7vNbo">Video</a> ] • [ <a href="images/live_shawe-taylor.jpg">Picture</a> ]
</li>
</ul>
<span class="image"><a href="images/live_panel.jpg"><img src="images/live_panel.jpg" height="604" width="806" alt="" data-position="center center" /></a></span>
</section>
<section id="five" class="wrapper style1 fade-up">
<div class="inner">
<h2>Organizers</h2>
<!-- <p>The workshop welcomes world-class experts on {quasi,PAC,∅}-Bayesian learning.</p> -->
<!-- <div class="features"> -->
<p align="center">
<a href="mailto:benjamin.guedj@inria.fr?subject=[Website]%20Your%20NIPS%20Workshop&cc=francis.bach@inria.fr,pascal.germain@inria.fr">Contact us!</a>
</p>
<table style="width:100%" align="center">
<tr align="center">
<td>
<span class="image"><img src="images/guedj.jpg" height="128" width="128" alt="" data-position="center center" /></span>
</td>
<td>
<span class="image"><img src="images/bach.jpg" height="128" width="128" alt="" data-position="center center" /></span>
</td>
<td>
<span class="image"><img src="images/germain.jpg" height="128" width="128" alt="" data-position="center center" /></span>
</td>
</tr>
<tr align="center">
<td>
<h3><a href="https://bguedj.github.io">Benjamin Guedj</a></h3>
</td>
<td>
<h3><a href="http://www.di.ens.fr/~fbach/">Francis Bach</a></h3>
</td>
<td>
<h3><a href="http://chercheurs.lille.inria.fr/pgermain/">Pascal Germain</a></h3>
</td>
</tr>
<tr align="center">
<td>
Researcher, Inria, France.
</td>
<td>
Senior Researcher, Inria, France.
</td>
<td>
Researcher, Inria, France.
</td>
</tr>
</table>
<h3>Sponsors</h3>
<span class="image"><a href="https://www.inria.fr/en/"><img src="images/inria.png" height="142" width="333" alt="" data-position="center center" /></a></span>
<span class="image"><a href="http://math.univ-lille1.fr/~cempi/index_eng.php"><img src="images/cempi.png" height="126" width="456" alt="" data-position="center center" /></a></span>
<span class="image"><a href="https://www.elementai.com"><img src="images/elementai.png" height="142" width="563" alt="" data-position="center center" /></a></span>
<h3>Endorsed by</h3>
<span class="image"><a href="https://bayesian.org"><img src="images/isba.jpg" height="88" width="270" alt="" data-position="center center" /></a></span>
</div>
</div>
</section>
</div>
<!-- Footer -->
<footer id="footer" class="wrapper style1-alt">
<div class="inner">
<ul class="menu">
<li>© Benjamin Guedj. All rights reserved.</li><li>Design: <a href="http://html5up.net">HTML5 UP</a></li>
</ul>
</div>
</footer>
<!-- Scripts -->
<script src="assets/js/jquery.min.js"></script>
<script src="assets/js/jquery.scrollex.min.js"></script>
<script src="assets/js/jquery.scrolly.min.js"></script>
<script src="assets/js/skel.min.js"></script>
<script src="assets/js/util.js"></script>
<!--[if lte IE 8]><script src="assets/js/ie/respond.min.js"></script><![endif]-->
<script src="assets/js/main.js"></script>
<!-- Google Analytics -->
<script>
(function(i,s,o,g,r,a,m){i['GoogleAnalyticsObject']=r;i[r]=i[r]||function(){
(i[r].q=i[r].q||[]).push(arguments)},i[r].l=1*new Date();a=s.createElement(o),
m=s.getElementsByTagName(o)[0];a.async=1;a.src=g;m.parentNode.insertBefore(a,m)
})(window,document,'script','https://www.google-analytics.com/analytics.js','ga');
ga('create', 'UA-81430854-1', 'auto');
ga('send', 'pageview');
</script>
</body>
</html>