Feature Selection using Stochastic Gates

Yutaro Yamada, Ofir Lindenbaum, Sahand Negahban, Yuval Kluger
Proceedings of the 37th International Conference on Machine Learning, PMLR 119:10648-10659, 2020.

Abstract

Feature selection problems have been extensively studied in the setting of linear estimation (e.g. LASSO), but less emphasis has been placed on feature selection for non-linear functions. In this study, we propose a method for feature selection in neural network estimation problems. The new procedure is based on probabilistic relaxation of the $\ell_0$ norm of features, or the count of the number of selected features. Our $\ell_0$-based regularization relies on a continuous relaxation of the Bernoulli distribution; such relaxation allows our model to learn the parameters of the approximate Bernoulli distributions via gradient descent. The proposed framework simultaneously learns either a nonlinear regression or classification function while selecting a small subset of features. We provide an information-theoretic justification for incorporating Bernoulli distribution into feature selection. Furthermore, we evaluate our method using synthetic and real-life data to demonstrate that our approach outperforms other commonly used methods in both predictive performance and feature selection.

Cite this Paper


BibTeX
@InProceedings{pmlr-v119-yamada20a, title = {Feature Selection using Stochastic Gates}, author = {Yamada, Yutaro and Lindenbaum, Ofir and Negahban, Sahand and Kluger, Yuval}, booktitle = {Proceedings of the 37th International Conference on Machine Learning}, pages = {10648--10659}, 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/yamada20a/yamada20a.pdf}, url = {http://proceedings.mlr.press/v119/yamada20a.html}, abstract = {Feature selection problems have been extensively studied in the setting of linear estimation (e.g. LASSO), but less emphasis has been placed on feature selection for non-linear functions. In this study, we propose a method for feature selection in neural network estimation problems. The new procedure is based on probabilistic relaxation of the $\ell_0$ norm of features, or the count of the number of selected features. Our $\ell_0$-based regularization relies on a continuous relaxation of the Bernoulli distribution; such relaxation allows our model to learn the parameters of the approximate Bernoulli distributions via gradient descent. The proposed framework simultaneously learns either a nonlinear regression or classification function while selecting a small subset of features. We provide an information-theoretic justification for incorporating Bernoulli distribution into feature selection. Furthermore, we evaluate our method using synthetic and real-life data to demonstrate that our approach outperforms other commonly used methods in both predictive performance and feature selection.} }
Endnote
%0 Conference Paper %T Feature Selection using Stochastic Gates %A Yutaro Yamada %A Ofir Lindenbaum %A Sahand Negahban %A Yuval Kluger %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-yamada20a %I PMLR %P 10648--10659 %U http://proceedings.mlr.press/v119/yamada20a.html %V 119 %X Feature selection problems have been extensively studied in the setting of linear estimation (e.g. LASSO), but less emphasis has been placed on feature selection for non-linear functions. In this study, we propose a method for feature selection in neural network estimation problems. The new procedure is based on probabilistic relaxation of the $\ell_0$ norm of features, or the count of the number of selected features. Our $\ell_0$-based regularization relies on a continuous relaxation of the Bernoulli distribution; such relaxation allows our model to learn the parameters of the approximate Bernoulli distributions via gradient descent. The proposed framework simultaneously learns either a nonlinear regression or classification function while selecting a small subset of features. We provide an information-theoretic justification for incorporating Bernoulli distribution into feature selection. Furthermore, we evaluate our method using synthetic and real-life data to demonstrate that our approach outperforms other commonly used methods in both predictive performance and feature selection.
APA
Yamada, Y., Lindenbaum, O., Negahban, S. & Kluger, Y.. (2020). Feature Selection using Stochastic Gates. Proceedings of the 37th International Conference on Machine Learning, in Proceedings of Machine Learning Research 119:10648-10659 Available from http://proceedings.mlr.press/v119/yamada20a.html.

Related Material