Mercurial > ift6266

annotate writeup/aistats_review_response.txt @ 642:8b1a0b9fecff

Find changesets by keywords (author, files, the commit message), revision number or hash, or revset expression.
made publication list shorter.
author Frederic Bastien <nouiz@nouiz.org>
date Mon, 21 Mar 2011 11:29:09 -0400
parents 49933073590c
children
rev   line source
616
b0cdd200b2bd added aistats_review_response.txt
Yoshua Bengio <bengioy@iro.umontreal.ca>
parents:
diff changeset
1
619
ea31fee25147 review response
Yoshua Bengio <bengioy@iro.umontreal.ca>
parents: 618
diff changeset
2 We thank the authors for their thoughtful comments. Please find our responses below.
616
b0cdd200b2bd added aistats_review_response.txt
Yoshua Bengio <bengioy@iro.umontreal.ca>
parents:
diff changeset
3
624
49933073590c added jmlr_review1.txt and jmlr_review2.txt
Yoshua Bengio <bengioy@iro.umontreal.ca>
parents: 623
diff changeset
4 * Comparisons with shallower networks, but using unsupervised pre-training. We have added those results to the paper. On the NIST test set, 62 classes,
49933073590c added jmlr_review1.txt and jmlr_review2.txt
Yoshua Bengio <bengioy@iro.umontreal.ca>
parents: 623
diff changeset
5 using NISTP to train (which gives the best results on NIST):
49933073590c added jmlr_review1.txt and jmlr_review2.txt
Yoshua Bengio <bengioy@iro.umontreal.ca>
parents: 623
diff changeset
6 MLP (1 hidden layer, no unsupervised pre-training): 24% error
49933073590c added jmlr_review1.txt and jmlr_review2.txt
Yoshua Bengio <bengioy@iro.umontreal.ca>
parents: 623
diff changeset
7 DA (1 hidden layer, unsupervised pre-training): 21% error
49933073590c added jmlr_review1.txt and jmlr_review2.txt
Yoshua Bengio <bengioy@iro.umontreal.ca>
parents: 623
diff changeset
8 SDA (2 hidden layers, unsupervised pre-training): 20% error
49933073590c added jmlr_review1.txt and jmlr_review2.txt
Yoshua Bengio <bengioy@iro.umontreal.ca>
parents: 623
diff changeset
9 SDA (3 hidden layers, unsupervised pre-training): 17% error
49933073590c added jmlr_review1.txt and jmlr_review2.txt
Yoshua Bengio <bengioy@iro.umontreal.ca>
parents: 623
diff changeset
10 Previous work in our group with very similar data (the InfiniteMNIST dataset were published in JMLR in 2010 "Why Does Unsupervised Pre-training Help Deep Learning?"). The results indeed show improvement when going from 1 to 2 and then 3 layers, even when using unsupervised pre-training (RBM or Denoising Auto-Encoder). The experiment helps to disentangle to some extent the effect of depth with the effect of unsupervised pre-training, and confirms that both are required to achieve the best results.
616
b0cdd200b2bd added aistats_review_response.txt
Yoshua Bengio <bengioy@iro.umontreal.ca>
parents:
diff changeset
11
624
49933073590c added jmlr_review1.txt and jmlr_review2.txt
Yoshua Bengio <bengioy@iro.umontreal.ca>
parents: 623
diff changeset
12 * Comparisons with SVMs. The main limitation of course is the size of the training set. One option is to use a non-linear SVM with a reduced training set, and the other is to use an online linear SVM. Another option is to project the input non-linearly in a high-dimensional but sparse representation and then use an online linear SVM. For this, we have thresholded input pixel gray levels and projected into the space of order-2 products. Results:
616
b0cdd200b2bd added aistats_review_response.txt
Yoshua Bengio <bengioy@iro.umontreal.ca>
parents:
diff changeset
13
b0cdd200b2bd added aistats_review_response.txt
Yoshua Bengio <bengioy@iro.umontreal.ca>
parents:
diff changeset
14 SVM type training set input online validation test set
b0cdd200b2bd added aistats_review_response.txt
Yoshua Bengio <bengioy@iro.umontreal.ca>
parents:
diff changeset
15 type / size features training set error error
b0cdd200b2bd added aistats_review_response.txt
Yoshua Bengio <bengioy@iro.umontreal.ca>
parents:
diff changeset
16 error
b0cdd200b2bd added aistats_review_response.txt
Yoshua Bengio <bengioy@iro.umontreal.ca>
parents:
diff changeset
17 Linear SVM, NIST, 651k, original, 36.62%, 34.41%, 42.26%
b0cdd200b2bd added aistats_review_response.txt
Yoshua Bengio <bengioy@iro.umontreal.ca>
parents:
diff changeset
18 Linear SVM, NIST, 651k, sparse quadratic, 30.96%, 28.00%, 41.28%
b0cdd200b2bd added aistats_review_response.txt
Yoshua Bengio <bengioy@iro.umontreal.ca>
parents:
diff changeset
19 Linear SVM, NISTP, 800k, original, 88.50%, 85.24%, 87.36%
b0cdd200b2bd added aistats_review_response.txt
Yoshua Bengio <bengioy@iro.umontreal.ca>
parents:
diff changeset
20 Linear SVM, NISTP, 800k, sparse quadratic, 81.76%, 83.69%, 85.56%
b0cdd200b2bd added aistats_review_response.txt
Yoshua Bengio <bengioy@iro.umontreal.ca>
parents:
diff changeset
21 RBF SVM, NISTP, 100k, original, 74.73%, 56.57%, 64.22%
b0cdd200b2bd added aistats_review_response.txt
Yoshua Bengio <bengioy@iro.umontreal.ca>
parents:
diff changeset
22
624
49933073590c added jmlr_review1.txt and jmlr_review2.txt
Yoshua Bengio <bengioy@iro.umontreal.ca>
parents: 623
diff changeset
23 The best results were obtained with the sparse quadratic input features, and training on the clean data (NIST) rather than the perturbed data (NISTP). A summary of the above results was added to the revised paper.
616
b0cdd200b2bd added aistats_review_response.txt
Yoshua Bengio <bengioy@iro.umontreal.ca>
parents:
diff changeset
24
b0cdd200b2bd added aistats_review_response.txt
Yoshua Bengio <bengioy@iro.umontreal.ca>
parents:
diff changeset
25
624
49933073590c added jmlr_review1.txt and jmlr_review2.txt
Yoshua Bengio <bengioy@iro.umontreal.ca>
parents: 623
diff changeset
26 * Using distorted characters as the corruption process of the Denoising Auto-Encoder (DAE). We had already performed preliminary experiments with this idea and results varied depending on the type of distortion, but did not improve on the original noise process. We believe that the DAE learns good features when the target to reconstruct is more likely than the corrupted input. concentrate. Hence distortions that are *plausible* in the input distribution (such as translation, rotation, scaling, etc.) are not very useful, whereas corruption due to a form of noise are useful. Consider also the symmetries involved: a translation is as likely to be to the right or to the left, so it is hard to predict.
616
b0cdd200b2bd added aistats_review_response.txt
Yoshua Bengio <bengioy@iro.umontreal.ca>
parents:
diff changeset
27
624
49933073590c added jmlr_review1.txt and jmlr_review2.txt
Yoshua Bengio <bengioy@iro.umontreal.ca>
parents: 623
diff changeset
28 * Human labeling: We controlled noise in the labelling process by (1) requiring AMT workers with a higher than normal average of accepted responses (>95%) on other tasks (2) discarding responses that were not complete (10 predictions) (3) discarding responses for which for which the time to predict was smaller than 3 seconds for NIST (the mean response time was 20 seconds) and 6 seconds seconds for NISTP (average response time of 45 seconds) (4) discarding responses which were obviously wrong (10 identical ones, or "12345..."). Overall, after such filtering, we kept approximately 95% of the AMT workers' responses. The above paragraph was added to the revision. We thank the reviewer for the suggestion about multi-stage questionnaires, we will definitely consider this as an option next time we perform this experiment. However, to be fair, if we were to do so, we should also consider the same multi-stage decision process for the machine learning algorithms as well.
618
14ba0120baff review response changes
Yoshua Bengio <bengioy@iro.umontreal.ca>
parents: 616
diff changeset
29
624
49933073590c added jmlr_review1.txt and jmlr_review2.txt
Yoshua Bengio <bengioy@iro.umontreal.ca>
parents: 623
diff changeset
30 * Size of labeled set: in our JMLR 2010 paper on deep learning (cited above, see fig. 11), we already verified the effect of number of labeled examples on the deep learners and shallow learners (with or without unsupervised pre-training). Basically (and somewhat surprisingly) the deep learners with unsupervised pre-training can take more advantage of a large amount of labeled examples, presumably because of the initialization effect and the effect does not disappear when the number of labeled examples increases. Similar results were obtained in the semi-supervised setting (Lee et al, NIPS2009). Adding the training curve in the self-taught settings of this AISTAT submission is a good idea, and we will have it for the final version.
616
b0cdd200b2bd added aistats_review_response.txt
Yoshua Bengio <bengioy@iro.umontreal.ca>
parents:
diff changeset
31