Data-Efficient Image Recognition with Contrastive Predictive Coding

Olivier Henaff
Proceedings of the 37th International Conference on Machine Learning, PMLR 119:4182-4192, 2020.

Abstract

Human observers can learn to recognize new categories of images from a handful of examples, yet doing so with artificial ones remains an open challenge. We hypothesize that data-efficient recognition is enabled by representations which make the variability in natural signals more predictable. We therefore revisit and improve Contrastive Predictive Coding, an unsupervised objective for learning such representations. This new implementation produces features which support state-of-the-art linear classification accuracy on the ImageNet dataset. When used as input for non-linear classification with deep neural networks, this representation allows us to use 2-5x less labels than classifiers trained directly on image pixels. Finally, this unsupervised representation substantially improves transfer learning to object detection on the PASCAL VOC dataset, surpassing fully supervised pre-trained ImageNet classifiers.

Cite this Paper

BibTeX
@InProceedings{pmlr-v119-henaff20a, title = {Data-Efficient Image Recognition with Contrastive Predictive Coding}, author = {Henaff, Olivier}, booktitle = {Proceedings of the 37th International Conference on Machine Learning}, pages = {4182--4192}, 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/henaff20a/henaff20a.pdf}, url = {https://proceedings.mlr.press/v119/henaff20a.html}, abstract = {Human observers can learn to recognize new categories of images from a handful of examples, yet doing so with artificial ones remains an open challenge. We hypothesize that data-efficient recognition is enabled by representations which make the variability in natural signals more predictable. We therefore revisit and improve Contrastive Predictive Coding, an unsupervised objective for learning such representations. This new implementation produces features which support state-of-the-art linear classification accuracy on the ImageNet dataset. When used as input for non-linear classification with deep neural networks, this representation allows us to use 2-5x less labels than classifiers trained directly on image pixels. Finally, this unsupervised representation substantially improves transfer learning to object detection on the PASCAL VOC dataset, surpassing fully supervised pre-trained ImageNet classifiers.} }
Endnote
%0 Conference Paper %T Data-Efficient Image Recognition with Contrastive Predictive Coding %A Olivier Henaff %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-henaff20a %I PMLR %P 4182--4192 %U https://proceedings.mlr.press/v119/henaff20a.html %V 119 %X Human observers can learn to recognize new categories of images from a handful of examples, yet doing so with artificial ones remains an open challenge. We hypothesize that data-efficient recognition is enabled by representations which make the variability in natural signals more predictable. We therefore revisit and improve Contrastive Predictive Coding, an unsupervised objective for learning such representations. This new implementation produces features which support state-of-the-art linear classification accuracy on the ImageNet dataset. When used as input for non-linear classification with deep neural networks, this representation allows us to use 2-5x less labels than classifiers trained directly on image pixels. Finally, this unsupervised representation substantially improves transfer learning to object detection on the PASCAL VOC dataset, surpassing fully supervised pre-trained ImageNet classifiers.
APA
Henaff, O.. (2020). Data-Efficient Image Recognition with Contrastive Predictive Coding. Proceedings of the 37th International Conference on Machine Learning, in Proceedings of Machine Learning Research 119:4182-4192 Available from https://proceedings.mlr.press/v119/henaff20a.html.

Related Material