Parameterized Rate-Distortion Stochastic Encoder

Quan Hoang, Trung Le, Dinh Phung
Proceedings of the 37th International Conference on Machine Learning, PMLR 119:4293-4303, 2020.

Abstract

We propose a novel gradient-based tractable approach for the Blahut-Arimoto (BA) algorithm to compute the rate-distortion function where the BA algorithm is fully parameterized. This results in a rich and flexible framework to learn a new class of stochastic encoders, termed PArameterized RAte-DIstortion Stochastic Encoder (PARADISE). The framework can be applied to a wide range of settings from semi-supervised, multi-task to supervised and robust learning. We show that the training objective of PARADISE can be seen as a form of regularization that helps improve generalization. With an emphasis on robust learning we further develop a novel posterior matching objective to encourage smoothness on the loss function and show that PARADISE can significantly improve interpretability as well as robustness to adversarial attacks on the CIFAR-10 and ImageNet datasets. In particular, on the CIFAR-10 dataset, our model reduces standard and adversarial error rates in comparison to the state-of-the-art by 50% and 41%, respectively without the expensive computational cost of adversarial training.

Cite this Paper


BibTeX
@InProceedings{pmlr-v119-hoang20c, title = {Parameterized Rate-Distortion Stochastic Encoder}, author = {Hoang, Quan and Le, Trung and Phung, Dinh}, booktitle = {Proceedings of the 37th International Conference on Machine Learning}, pages = {4293--4303}, 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/hoang20c/hoang20c.pdf}, url = {http://proceedings.mlr.press/v119/hoang20c.html}, abstract = {We propose a novel gradient-based tractable approach for the Blahut-Arimoto (BA) algorithm to compute the rate-distortion function where the BA algorithm is fully parameterized. This results in a rich and flexible framework to learn a new class of stochastic encoders, termed PArameterized RAte-DIstortion Stochastic Encoder (PARADISE). The framework can be applied to a wide range of settings from semi-supervised, multi-task to supervised and robust learning. We show that the training objective of PARADISE can be seen as a form of regularization that helps improve generalization. With an emphasis on robust learning we further develop a novel posterior matching objective to encourage smoothness on the loss function and show that PARADISE can significantly improve interpretability as well as robustness to adversarial attacks on the CIFAR-10 and ImageNet datasets. In particular, on the CIFAR-10 dataset, our model reduces standard and adversarial error rates in comparison to the state-of-the-art by 50% and 41%, respectively without the expensive computational cost of adversarial training.} }
Endnote
%0 Conference Paper %T Parameterized Rate-Distortion Stochastic Encoder %A Quan Hoang %A Trung Le %A Dinh Phung %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-hoang20c %I PMLR %P 4293--4303 %U http://proceedings.mlr.press/v119/hoang20c.html %V 119 %X We propose a novel gradient-based tractable approach for the Blahut-Arimoto (BA) algorithm to compute the rate-distortion function where the BA algorithm is fully parameterized. This results in a rich and flexible framework to learn a new class of stochastic encoders, termed PArameterized RAte-DIstortion Stochastic Encoder (PARADISE). The framework can be applied to a wide range of settings from semi-supervised, multi-task to supervised and robust learning. We show that the training objective of PARADISE can be seen as a form of regularization that helps improve generalization. With an emphasis on robust learning we further develop a novel posterior matching objective to encourage smoothness on the loss function and show that PARADISE can significantly improve interpretability as well as robustness to adversarial attacks on the CIFAR-10 and ImageNet datasets. In particular, on the CIFAR-10 dataset, our model reduces standard and adversarial error rates in comparison to the state-of-the-art by 50% and 41%, respectively without the expensive computational cost of adversarial training.
APA
Hoang, Q., Le, T. & Phung, D.. (2020). Parameterized Rate-Distortion Stochastic Encoder. Proceedings of the 37th International Conference on Machine Learning, in Proceedings of Machine Learning Research 119:4293-4303 Available from http://proceedings.mlr.press/v119/hoang20c.html.

Related Material