Learning Discrete Structured Representations by Adversarially Maximizing Mutual Information

Karl Stratos, Sam Wiseman
Proceedings of the 37th International Conference on Machine Learning, PMLR 119:9144-9154, 2020.

Abstract

We propose learning discrete structured representations from unlabeled data by maximizing the mutual information between a structured latent variable and a target variable. Calculating mutual information is intractable in this setting. Our key technical contribution is an adversarial objective that can be used to tractably estimate mutual information assuming only the feasibility of cross entropy calculation. We develop a concrete realization of this general formulation with Markov distributions over binary encodings. We report critical and unexpected findings on practical aspects of the objective such as the choice of variational priors. We apply our model on document hashing and show that it outperforms current best baselines based on discrete and vector quantized variational autoencoders. It also yields highly compressed interpretable representations.

Cite this Paper

BibTeX
@InProceedings{pmlr-v119-stratos20a, title = {Learning Discrete Structured Representations by Adversarially Maximizing Mutual Information}, author = {Stratos, Karl and Wiseman, Sam}, booktitle = {Proceedings of the 37th International Conference on Machine Learning}, pages = {9144--9154}, 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/stratos20a/stratos20a.pdf}, url = {https://proceedings.mlr.press/v119/stratos20a.html}, abstract = {We propose learning discrete structured representations from unlabeled data by maximizing the mutual information between a structured latent variable and a target variable. Calculating mutual information is intractable in this setting. Our key technical contribution is an adversarial objective that can be used to tractably estimate mutual information assuming only the feasibility of cross entropy calculation. We develop a concrete realization of this general formulation with Markov distributions over binary encodings. We report critical and unexpected findings on practical aspects of the objective such as the choice of variational priors. We apply our model on document hashing and show that it outperforms current best baselines based on discrete and vector quantized variational autoencoders. It also yields highly compressed interpretable representations.} }
Endnote
%0 Conference Paper %T Learning Discrete Structured Representations by Adversarially Maximizing Mutual Information %A Karl Stratos %A Sam Wiseman %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-stratos20a %I PMLR %P 9144--9154 %U https://proceedings.mlr.press/v119/stratos20a.html %V 119 %X We propose learning discrete structured representations from unlabeled data by maximizing the mutual information between a structured latent variable and a target variable. Calculating mutual information is intractable in this setting. Our key technical contribution is an adversarial objective that can be used to tractably estimate mutual information assuming only the feasibility of cross entropy calculation. We develop a concrete realization of this general formulation with Markov distributions over binary encodings. We report critical and unexpected findings on practical aspects of the objective such as the choice of variational priors. We apply our model on document hashing and show that it outperforms current best baselines based on discrete and vector quantized variational autoencoders. It also yields highly compressed interpretable representations.
APA
Stratos, K. & Wiseman, S.. (2020). Learning Discrete Structured Representations by Adversarially Maximizing Mutual Information. Proceedings of the 37th International Conference on Machine Learning, in Proceedings of Machine Learning Research 119:9144-9154 Available from https://proceedings.mlr.press/v119/stratos20a.html.

Related Material