Mercurial > ift6266

comparison writeup/ReviewsAISTATSfinal.html @ 643:24d9819a810f

Find changesets by keywords (author, files, the commit message), revision number or hash, or revset expression.
reviews aistats finales
author Yoshua Bengio <bengioy@iro.umontreal.ca>
date Thu, 24 Mar 2011 17:04:38 -0400
parents
children
comparison
equal deleted inserted replaced
639:507cb92d8e15 643:24d9819a810f
1 <!DOCTYPE html PUBLIC "-//W3C//DTD XHTML 1.0 Transitional//EN" "http://www.w3.org/TR/xhtml1/DTD/xhtml1-transitional.dtd">
2 <!-- saved from url=(0096)https://cmt.research.microsoft.com/AIS2011/Protected/Author/ViewReviewsForPaper.aspx?paperId=126 -->
3 <html xmlns="http://www.w3.org/1999/xhtml"><head><meta http-equiv="Content-Type" content="text/html; charset=UTF-8"><title>
4 Reviews For Paper
5 </title>
6 <style>
7 #header
8 {
9 width: 100%;
10 font-size: small;
11 background-color:#F7F7F7;
12 }
13 .printThemeText
14 {
15 font-size:small;
16 }
17 .printThemeTable td
18 {
19 vertical-align:top;
20 }
21 .printThemeGrid th
22 {
23 color:white;
24 background:#5D7B9D;
25 font-weight:bold;
26 }
27 .printThemeGrid
28 {
29 border-collapse:collapse;
30 }
31 .printThemeGrid td, .printThemeGrid th
32 {
33 border:solid 1px #D6D3CE;
34 padding:4px 4px 4px 4px;
35 }
36 .printThemeGrid .row
37 {
38 background-color:#F7F6F3;
39 color:#333333;
40 vertical-align:top;
41 }
42 .printThemeGrid .altrow
43 {
44 background-color:White;
45 color:#284775;
46 vertical-align:top;
47 }
48 .cellprompt
49 {
50 font-weight:bold;
51 white-space:nowrap;
52 width:100px;
53 }
54 .paperHeader
55 {
56 background-color:#dee3e7;
57 margin:5px 5px 15px 0px;
58 width:99%;
59 font-family:Verdana;
60 font-size:medium;
61 font-weight:bold;
62 }
63 .sectionHeader
64 {
65 background-color:#dee3e7;
66 padding:5px 5px 5px 0px;
67 width:99%;
68 text-decoration:underline;
69 font-family:Verdana;
70 font-size:small;
71 font-weight:bold;
72 }
73 .underlineheader
74 {
75 text-decoration:underline;
76 font-weight:bold;
77 padding:5px 0px;
78 }
79 .response
80 {
81 padding:5px 0px;
82 }
83 .reviewerlabel
84 {
85 padding-right:20px;
86 }
87 .pageTitle
88 {
89 background-color:#dee3e7;
90 padding:5px 5px 5px 5px;
91 margin-top:10px;
92 width:99%;
93 font-family:Verdana;
94 font-size:medium;
95 font-weight:bold;
96 }
97 .submissionDetailsView
98 {
99 }
100 .submissionDetailsView tr
101 {
102 vertical-align:top;
103 }
104 .submissionDetailsView td.prompt
105 {
106 font-weight:bold;
107 }
108 .submissionDetailsView tr.sectionSeparator
109 {
110
111 }
112 .submissionDetailsView tr.sectionSeparator td
113 {
114 background-color:#dee3e7;
115 padding:5px 5px 5px 5px;
116 font-family:Verdana;
117 font-size:small;
118 font-weight:bold;
119 color:Navy;
120 }
121 </style>
122 </head>
123 <body>
124 <form name="aspnetForm" method="post" action="./ReviewsAISTATSfinal_files/ReviewsAISTATSfinal.html" id="aspnetForm">
125 <div>
126 <input type="hidden" name="__VIEWSTATE" id="__VIEWSTATE" value="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">
127 </div>
128
129 <table id="header">
130 <tbody><tr>
131 <td><img src="./ReviewsAISTATSfinal_files/conferencelogo.gif"></td>
132 <td width="100%"><a href="http://www.aistats.org/">AI &amp; Statistics 2011 </a><br><b>Fourteenth International Conference on Artificial Intelligence and Statistics </b><br>April 11-13, 2011<br>Ft. Lauderdale, FL<br>USA</td>
133 </tr>
134 </tbody></table>
135 <table id="content"><tbody><tr><td class="contentBorder">&nbsp;</td><td class="contentContainer">
136 <span id="ctl00_cph_Label4" style="font-size:Small;font-weight:bold;">Reviews For Paper</span>
137 <span id="ctl00_cph_lblErrorMessage" class="error" style="font-size:Small;"></span>
138 <div id="ctl00_cph_pnlReviews">
139
140 <span style="font-size:Small;">
141 <table class="nicetable2" style="text-align:left; width: 100%;">
142
143 <tbody><tr>
144 <td width="100px"><b>Paper ID</b></td>
145 <td><span id="ctl00_cph_infoSubmission_lblPaperId" style="font-size:Small;">126</span></td>
146 </tr>
147 <tr>
148 <td><b>Title</b></td>
149 <td><span id="ctl00_cph_infoSubmission_lblPaperTitle" style="font-size:Small;">Deep Learners Benefit More from Out-of-Distribution Examples</span></td>
150 </tr>
151
152
153
154
155
156 </tbody></table></span>
157
158
159 <hr>
160 <table>
161 <tbody><tr>
162 <td>
163 <span id="ctl00_cph_gvReviews_ctl00_Label2" style="font-size:Small;font-weight:bold;">Masked Reviewer ID:</span>
164 </td>
165 <td>
166 <span id="ctl00_cph_gvReviews_ctl00_Label1" style="font-size:Small;">Assigned_Reviewer_2</span>
167 </td>
168 </tr>
169 <tr>
170 <td>
171 <span id="ctl00_cph_gvReviews_ctl00_Label3" style="font-size:Small;font-weight:bold;">Review:</span>
172 </td>
173 <td>
174 </td>
175 </tr>
176 </tbody></table>
177 <div>
178 <table cellspacing="0" cellpadding="4" rules="all" border="1" style="color:#333333;border-width:1px;border-style:None;font-family:Verdana;font-size:Small;border-collapse:collapse;">
179 <tbody><tr style="color:White;background-color:#5D7B9D;font-weight:bold;">
180 <th scope="col">Question</th><th scope="col">&nbsp;</th>
181 </tr><tr style="color:#333333;background-color:#F7F6F3;">
182 <td style="width:20%;">Overall rating: please synthesize your answers to other questions into an overall recommendation. Please take into account tradeoffs (an increase in one measure may compensate for a decrease in another), and describe the tradeoffs in the detailed comments.</td><td style="width:80%;">
183 Very good: suggest accept
184 </td>
185 </tr><tr style="color:#284775;background-color:White;">
186 <td style="width:20%;">Technical quality: is all included material presented clearly and correctly?</td><td style="width:80%;">
187 Good
188 </td>
189 </tr><tr style="color:#333333;background-color:#F7F6F3;">
190 <td style="width:20%;">Originality: how much new work is represented in this paper, beyond previous conference/journal papers?</td><td style="width:80%;">
191 Substantial new material
192 </td>
193 </tr><tr style="color:#284775;background-color:White;">
194 <td style="width:20%;">Interest and significance: would the paper's goal, if completely solved, represent a substantial advance for the AISTATS community?</td><td style="width:80%;">
195 Significant
196 </td>
197 </tr><tr style="color:#333333;background-color:#F7F6F3;">
198 <td style="width:20%;">Thoroughness: to what degree does the paper support its conclusions through experimental comparisons, theorems, etc.?</td><td style="width:80%;">
199 Thorough
200 </td>
201 </tr><tr style="color:#284775;background-color:White;">
202 <td style="width:20%;">Creativity: to what degree does the paper represent a novel way of setting up a problem or an unusual approach to solving it?</td><td style="width:80%;">
203 Most content represents application of known ideas
204 </td>
205 </tr><tr style="color:#333333;background-color:#F7F6F3;">
206 <td style="width:20%;">Detailed Comments</td><td style="width:80%;">
207 This paper shows that deep networks benefit more from out-of-distribution examples than shallower architectures on a large scale character recognition experiment. A thorough empirical validation shows that deep nets produce better discrimination (than shallower nets) when trained with distorted characters and when trained on multiple tasks.
208 <br>Although the methods used are already well established in the community, these results are significant and provide new insights on the representational power of this class of methods.
209 <br>
210 <br>COMMENTS AFTER READING SECOND VERSION
211 <br>The paper is ready for publication.
212 <br>
213 <br>- Minor comment: it would be helpful to add the functional form of the encoder and decoder function of the DAE. The reference to [30] (DAE is equivalent to Gaussian RBM trained with Score Matching) might not be so relevant if the authors use different kinds of encoder and decoder functions (in particular, is the reconstruction squashed through a logistic as usually done to model binary discrete variables?, are the weight matrices symmetric?)
214 </td>
215 </tr>
216 </tbody></table>
217 </div>
218
219
220 <hr>
221 <table>
222 <tbody><tr>
223 <td>
224 <span id="ctl00_cph_gvReviews_ctl01_Label2" style="font-size:Small;font-weight:bold;">Masked Reviewer ID:</span>
225 </td>
226 <td>
227 <span id="ctl00_cph_gvReviews_ctl01_Label1" style="font-size:Small;">Assigned_Reviewer_3</span>
228 </td>
229 </tr>
230 <tr>
231 <td>
232 <span id="ctl00_cph_gvReviews_ctl01_Label3" style="font-size:Small;font-weight:bold;">Review:</span>
233 </td>
234 <td>
235 </td>
236 </tr>
237 </tbody></table>
238 <div>
239 <table cellspacing="0" cellpadding="4" rules="all" border="1" style="color:#333333;border-width:1px;border-style:None;font-family:Verdana;font-size:Small;border-collapse:collapse;">
240 <tbody><tr style="color:White;background-color:#5D7B9D;font-weight:bold;">
241 <th scope="col">Question</th><th scope="col">&nbsp;</th>
242 </tr><tr style="color:#333333;background-color:#F7F6F3;">
243 <td style="width:20%;">Overall rating: please synthesize your answers to other questions into an overall recommendation. Please take into account tradeoffs (an increase in one measure may compensate for a decrease in another), and describe the tradeoffs in the detailed comments.</td><td style="width:80%;">
244 Very good: suggest accept
245 </td>
246 </tr><tr style="color:#284775;background-color:White;">
247 <td style="width:20%;">Technical quality: is all included material presented clearly and correctly?</td><td style="width:80%;">
248 Very good
249 </td>
250 </tr><tr style="color:#333333;background-color:#F7F6F3;">
251 <td style="width:20%;">Originality: how much new work is represented in this paper, beyond previous conference/journal papers?</td><td style="width:80%;">
252 Substantial new material
253 </td>
254 </tr><tr style="color:#284775;background-color:White;">
255 <td style="width:20%;">Interest and significance: would the paper's goal, if completely solved, represent a substantial advance for the AISTATS community?</td><td style="width:80%;">
256 Significant
257 </td>
258 </tr><tr style="color:#333333;background-color:#F7F6F3;">
259 <td style="width:20%;">Thoroughness: to what degree does the paper support its conclusions through experimental comparisons, theorems, etc.?</td><td style="width:80%;">
260 Thorough
261 </td>
262 </tr><tr style="color:#284775;background-color:White;">
263 <td style="width:20%;">Creativity: to what degree does the paper represent a novel way of setting up a problem or an unusual approach to solving it?</td><td style="width:80%;">
264 Most content represents novel approaches
265 </td>
266 </tr><tr style="color:#333333;background-color:#F7F6F3;">
267 <td style="width:20%;">Detailed Comments</td><td style="width:80%;">
268 This paper claims that using out-of-distribution examples can be more helpful in training deep architectures than shallow architectures. In order to test this hypothesis, the paper develops extensive transformations for image patches (i.e., images of handwritten characters) to generate a large-scale dataset of perturbed images. These out-of-distribution examples are trained using MLPs and stacked denoising auto-encoders (SDAs). In the experiments, the paper shows that SDAs outperform MLPs, achieving human-level performance for NIST dataset. The paper also provides two interesting experiments showing that: (1) SDAs can benefit from training perturbed data, even when testing on clean data; (2) SDAs can significantly benefit from multi-task learning.
269 <br>
270 <br>
271 <br>Questions, comments, and suggestions:
272 <br>1. Regarding the human labeling, I have some concerns about labeling noise/biases due to AMT. How were the anomalies in labeling or outliers controlled? Was there any procedure to minimize labeling noise/biases or to ensure that human labelers tried their best (e.g., filtering out random guesses or encouraging the labelers to consider all possibilities carefully before providing premature guesses)? For example, multi-stage questionnaires (e.g., asking "characters/digits", "uppercase/lowercase", then choosing one out of 10 digits, or 26 characters) might significantly reduce labeling noise/biases, rather than showing 62 candidate answers simultaneously.
273 <br>
274 <br>2. It seems that the paper fixed the number of hidden layers as three. Despite good performance of the proposed architecture, it is somewhat unclear whether the benefit comes mainly from deep architecture or the use of denoising auto-encoders.
275 <br>
276 <br>Therefore, it will be more interesting to see the effect of the number of layers and other pre-training methods (e.g., RBMs or auto-encoders). This experiment will clarify where the benefit comes from (i.e., deep architecture vs. pre-training modules) and provide more insights about the results.
277 <br>
278 <br>3. The paper briefly mentioned about the use of libSVM, but it will be useful to compare against the results using online SVM (e.g., PEGASOS).
279 <br>
280 <br>4. The paper also talks about the effect of large labeled data in self-taught learning setting. To strengthen the claim, it will be helpful to show the test accuracy as a function of number of labeled examples.
281 <br>
282 <br>Overall, the paper is clearly written, and it provides interesting experiments on large scale datasets, addressing a number of interesting questions related to deep learning and multi-task learning. Furthermore, this work can provide a new large scale benchmark dataset (beyond MNIST) for deep learning and machine learning research.
283 <br>
284 </td>
285 </tr>
286 </tbody></table>
287 </div>
288
289
290 <hr>
291 <table>
292 <tbody><tr>
293 <td>
294 <span id="ctl00_cph_gvReviews_ctl02_Label2" style="font-size:Small;font-weight:bold;">Masked Reviewer ID:</span>
295 </td>
296 <td>
297 <span id="ctl00_cph_gvReviews_ctl02_Label1" style="font-size:Small;">Assigned_Reviewer_4</span>
298 </td>
299 </tr>
300 <tr>
301 <td>
302 <span id="ctl00_cph_gvReviews_ctl02_Label3" style="font-size:Small;font-weight:bold;">Review:</span>
303 </td>
304 <td>
305 </td>
306 </tr>
307 </tbody></table>
308 <div>
309 <table cellspacing="0" cellpadding="4" rules="all" border="1" style="color:#333333;border-width:1px;border-style:None;font-family:Verdana;font-size:Small;border-collapse:collapse;">
310 <tbody><tr style="color:White;background-color:#5D7B9D;font-weight:bold;">
311 <th scope="col">Question</th><th scope="col">&nbsp;</th>
312 </tr><tr style="color:#333333;background-color:#F7F6F3;">
313 <td style="width:20%;">Overall rating: please synthesize your answers to other questions into an overall recommendation. Please take into account tradeoffs (an increase in one measure may compensate for a decrease in another), and describe the tradeoffs in the detailed comments.</td><td style="width:80%;">
314 Good: suggest accept
315 </td>
316 </tr><tr style="color:#284775;background-color:White;">
317 <td style="width:20%;">Technical quality: is all included material presented clearly and correctly?</td><td style="width:80%;">
318 Very good
319 </td>
320 </tr><tr style="color:#333333;background-color:#F7F6F3;">
321 <td style="width:20%;">Originality: how much new work is represented in this paper, beyond previous conference/journal papers?</td><td style="width:80%;">
322 Substantial new material
323 </td>
324 </tr><tr style="color:#284775;background-color:White;">
325 <td style="width:20%;">Interest and significance: would the paper's goal, if completely solved, represent a substantial advance for the AISTATS community?</td><td style="width:80%;">
326 Significant
327 </td>
328 </tr><tr style="color:#333333;background-color:#F7F6F3;">
329 <td style="width:20%;">Thoroughness: to what degree does the paper support its conclusions through experimental comparisons, theorems, etc.?</td><td style="width:80%;">
330 Thorough
331 </td>
332 </tr><tr style="color:#284775;background-color:White;">
333 <td style="width:20%;">Creativity: to what degree does the paper represent a novel way of setting up a problem or an unusual approach to solving it?</td><td style="width:80%;">
334 Most content represents application of known ideas
335 </td>
336 </tr><tr style="color:#333333;background-color:#F7F6F3;">
337 <td style="width:20%;">Detailed Comments</td><td style="width:80%;">
338 The paper demonstrates that deeper networks benefit from "related data" more than shallow ones, and is fairly well written. Along the way, it constructs a very powerful character recognition system.
339 <br>
340 <br>I have several minor suggestions that would improve the paper.
341 <br>
342 <br>First, the "human-level performance", as obtained by the Mechanical
343 <br>Turk, likely ignores the intrinsic noise and sloppiness of the human
344 <br>annotators. That is, the annotators may occasionally select an incorrect
345 <br>label having recognized the image correctly. This effect can be measured by
346 <br>creating a very clean dataset (so the true error rate is zero), and to
347 <br>have it labelled by MT. The error rate is likely to be greater than
348 <br>zero, which should be taken into the human-performance estimation.
349 <br>
350 <br>Second, the experiments convincingly showed that deep SDA nets do
351 <br>quite well and benefit from more related data. However, recent work
352 <br>[http://arxiv.org/abs/1003.0358] has shown that very deep neural
353 <br>networks are very effective in the regime explored precisely in this
354 <br>paper, even without pretraining of any kind. Therefore, it would help
355 <br>to redo these experiments with an SDA with 6 or
356 <br>even 8 layers (or more), and to train the deep networks with no
357 <br>pretraining but with a careful random initialization that
358 <br>uses the correct scale at each layer.
359 <br>
360 <br>Lastly, there are no MNIST results. It is important to compare on
361 <br>MNIST because there are many careful results on MNIST.
362 <br>
363 <br>
364 <br>
365 </td>
366 </tr>
367 </tbody></table>
368 </div>
369
370
371 <br>
372 <br>
373
374 </div>
375 </td><td class="contentBorder">&nbsp;</td></tr></tbody></table>
376 </form>
377
378
379 </body></html>