Safe Deep Semi-Supervised Learning for Unseen-Class Unlabeled Data

Lan-Zhe Guo, Zhen-Yu Zhang, Yuan Jiang, Yu-Feng Li, Zhi-Hua Zhou
Proceedings of the 37th International Conference on Machine Learning, PMLR 119:3897-3906, 2020.

Abstract

Deep semi-supervised learning (SSL) has been recently shown very effectively. However, its performance is seriously decreased when the class distribution is mismatched, among which a common situation is that unlabeled data contains some classes not seen in the labeled data. Efforts on this issue remain to be limited. This paper proposes a simple and effective safe deep SSL method to alleviate the harm caused by it. In theory, the result learned from the new method is never worse than learning from merely labeled data, and it is theoretically guaranteed that its generalization approaches the optimal in the order $O(\sqrt{d\ln(n)/n})$, even faster than the convergence rate in supervised learning associated with massive parameters. In the experiment of benchmark data, unlike the existing deep SSL methods which are no longer as good as supervised learning in 40% of unseen-class unlabeled data, the new method can still achieve performance gain in more than 60% of unseen-class unlabeled data. Moreover, the proposal is suitable for many deep SSL algorithms and can be easily extended to handle other cases of class distribution mismatch.

Cite this Paper


BibTeX
@InProceedings{pmlr-v119-guo20i, title = {Safe Deep Semi-Supervised Learning for Unseen-Class Unlabeled Data}, author = {Guo, Lan-Zhe and Zhang, Zhen-Yu and Jiang, Yuan and Li, Yu-Feng and Zhou, Zhi-Hua}, booktitle = {Proceedings of the 37th International Conference on Machine Learning}, pages = {3897--3906}, year = {2020}, editor = {III, Hal Daumé and Singh, Aarti}, volume = {119}, series = {Proceedings of Machine Learning Research}, month = {13--18 Jul}, publisher = {PMLR}, pdf = {http://proceedings.mlr.press/v119/guo20i/guo20i.pdf}, url = {http://proceedings.mlr.press/v119/guo20i.html}, abstract = {Deep semi-supervised learning (SSL) has been recently shown very effectively. However, its performance is seriously decreased when the class distribution is mismatched, among which a common situation is that unlabeled data contains some classes not seen in the labeled data. Efforts on this issue remain to be limited. This paper proposes a simple and effective safe deep SSL method to alleviate the harm caused by it. In theory, the result learned from the new method is never worse than learning from merely labeled data, and it is theoretically guaranteed that its generalization approaches the optimal in the order $O(\sqrt{d\ln(n)/n})$, even faster than the convergence rate in supervised learning associated with massive parameters. In the experiment of benchmark data, unlike the existing deep SSL methods which are no longer as good as supervised learning in 40% of unseen-class unlabeled data, the new method can still achieve performance gain in more than 60% of unseen-class unlabeled data. Moreover, the proposal is suitable for many deep SSL algorithms and can be easily extended to handle other cases of class distribution mismatch.} }
Endnote
%0 Conference Paper %T Safe Deep Semi-Supervised Learning for Unseen-Class Unlabeled Data %A Lan-Zhe Guo %A Zhen-Yu Zhang %A Yuan Jiang %A Yu-Feng Li %A Zhi-Hua Zhou %B Proceedings of the 37th International Conference on Machine Learning %C Proceedings of Machine Learning Research %D 2020 %E Hal Daumé III %E Aarti Singh %F pmlr-v119-guo20i %I PMLR %P 3897--3906 %U http://proceedings.mlr.press/v119/guo20i.html %V 119 %X Deep semi-supervised learning (SSL) has been recently shown very effectively. However, its performance is seriously decreased when the class distribution is mismatched, among which a common situation is that unlabeled data contains some classes not seen in the labeled data. Efforts on this issue remain to be limited. This paper proposes a simple and effective safe deep SSL method to alleviate the harm caused by it. In theory, the result learned from the new method is never worse than learning from merely labeled data, and it is theoretically guaranteed that its generalization approaches the optimal in the order $O(\sqrt{d\ln(n)/n})$, even faster than the convergence rate in supervised learning associated with massive parameters. In the experiment of benchmark data, unlike the existing deep SSL methods which are no longer as good as supervised learning in 40% of unseen-class unlabeled data, the new method can still achieve performance gain in more than 60% of unseen-class unlabeled data. Moreover, the proposal is suitable for many deep SSL algorithms and can be easily extended to handle other cases of class distribution mismatch.
APA
Guo, L., Zhang, Z., Jiang, Y., Li, Y. & Zhou, Z.. (2020). Safe Deep Semi-Supervised Learning for Unseen-Class Unlabeled Data. Proceedings of the 37th International Conference on Machine Learning, in Proceedings of Machine Learning Research 119:3897-3906 Available from http://proceedings.mlr.press/v119/guo20i.html.

Related Material