Time-Consistent Self-Supervision for Semi-Supervised Learning

Tianyi Zhou, Shengjie Wang, Jeff Bilmes
Proceedings of the 37th International Conference on Machine Learning, PMLR 119:11523-11533, 2020.

Abstract

Semi-supervised learning (SSL) leverages unlabeled data when training a model with insufficient labeled data. A common strategy for SSL is to enforce the consistency of model outputs between similar samples, e.g., neighbors or data augmentations of the same sample. However, model outputs can vary dramatically on unlabeled data over different training stages, e.g., when using large learning rates. This can introduce harmful noises and inconsistent objectives over time that may lead to concept drift and catastrophic forgetting. In this paper, we study the dynamics of neural net outputs in SSL and show that selecting and using first the unlabeled samples with more consistent outputs over the course of training (i.e., "time-consistency") can improve the final test accuracy and save computation. Under the time-consistent data selection, we design an SSL objective composed of two self-supervised losses, i.e., a consistency loss between a sample and its augmentation, and a contrastive loss encouraging different samples to have different outputs. Our approach achieves SOTA on several SSL benchmarks with much fewer computations.

Cite this Paper


BibTeX
@InProceedings{pmlr-v119-zhou20d, title = {Time-Consistent Self-Supervision for Semi-Supervised Learning}, author = {Zhou, Tianyi and Wang, Shengjie and Bilmes, Jeff}, booktitle = {Proceedings of the 37th International Conference on Machine Learning}, pages = {11523--11533}, 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/zhou20d/zhou20d.pdf}, url = {https://proceedings.mlr.press/v119/zhou20d.html}, abstract = {Semi-supervised learning (SSL) leverages unlabeled data when training a model with insufficient labeled data. A common strategy for SSL is to enforce the consistency of model outputs between similar samples, e.g., neighbors or data augmentations of the same sample. However, model outputs can vary dramatically on unlabeled data over different training stages, e.g., when using large learning rates. This can introduce harmful noises and inconsistent objectives over time that may lead to concept drift and catastrophic forgetting. In this paper, we study the dynamics of neural net outputs in SSL and show that selecting and using first the unlabeled samples with more consistent outputs over the course of training (i.e., "time-consistency") can improve the final test accuracy and save computation. Under the time-consistent data selection, we design an SSL objective composed of two self-supervised losses, i.e., a consistency loss between a sample and its augmentation, and a contrastive loss encouraging different samples to have different outputs. Our approach achieves SOTA on several SSL benchmarks with much fewer computations.} }
Endnote
%0 Conference Paper %T Time-Consistent Self-Supervision for Semi-Supervised Learning %A Tianyi Zhou %A Shengjie Wang %A Jeff Bilmes %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-zhou20d %I PMLR %P 11523--11533 %U https://proceedings.mlr.press/v119/zhou20d.html %V 119 %X Semi-supervised learning (SSL) leverages unlabeled data when training a model with insufficient labeled data. A common strategy for SSL is to enforce the consistency of model outputs between similar samples, e.g., neighbors or data augmentations of the same sample. However, model outputs can vary dramatically on unlabeled data over different training stages, e.g., when using large learning rates. This can introduce harmful noises and inconsistent objectives over time that may lead to concept drift and catastrophic forgetting. In this paper, we study the dynamics of neural net outputs in SSL and show that selecting and using first the unlabeled samples with more consistent outputs over the course of training (i.e., "time-consistency") can improve the final test accuracy and save computation. Under the time-consistent data selection, we design an SSL objective composed of two self-supervised losses, i.e., a consistency loss between a sample and its augmentation, and a contrastive loss encouraging different samples to have different outputs. Our approach achieves SOTA on several SSL benchmarks with much fewer computations.
APA
Zhou, T., Wang, S. & Bilmes, J.. (2020). Time-Consistent Self-Supervision for Semi-Supervised Learning. Proceedings of the 37th International Conference on Machine Learning, in Proceedings of Machine Learning Research 119:11523-11533 Available from https://proceedings.mlr.press/v119/zhou20d.html.

Related Material