On the Implicit Bias of Dropout

Poorya Mianjy, Raman Arora, Rene Vidal
Proceedings of the 35th International Conference on Machine Learning, PMLR 80:3540-3548, 2018.

Abstract

Algorithmic approaches endow deep learning systems with implicit bias that helps them generalize even in over-parametrized settings. In this paper, we focus on understanding such a bias induced in learning through dropout, a popular technique to avoid overfitting in deep learning. For single hidden-layer linear neural networks, we show that dropout tends to make the norm of incoming/outgoing weight vectors of all the hidden nodes equal. In addition, we provide a complete characterization of the optimization landscape induced by dropout.

Cite this Paper

BibTeX
@InProceedings{pmlr-v80-mianjy18b, title = {On the Implicit Bias of Dropout}, author = {Mianjy, Poorya and Arora, Raman and Vidal, Rene}, booktitle = {Proceedings of the 35th International Conference on Machine Learning}, pages = {3540--3548}, year = {2018}, editor = {Dy, Jennifer and Krause, Andreas}, volume = {80}, series = {Proceedings of Machine Learning Research}, month = {10--15 Jul}, publisher = {PMLR}, pdf = {http://proceedings.mlr.press/v80/mianjy18b/mianjy18b.pdf}, url = {https://proceedings.mlr.press/v80/mianjy18b.html}, abstract = {Algorithmic approaches endow deep learning systems with implicit bias that helps them generalize even in over-parametrized settings. In this paper, we focus on understanding such a bias induced in learning through dropout, a popular technique to avoid overfitting in deep learning. For single hidden-layer linear neural networks, we show that dropout tends to make the norm of incoming/outgoing weight vectors of all the hidden nodes equal. In addition, we provide a complete characterization of the optimization landscape induced by dropout.} }
Endnote
%0 Conference Paper %T On the Implicit Bias of Dropout %A Poorya Mianjy %A Raman Arora %A Rene Vidal %B Proceedings of the 35th International Conference on Machine Learning %C Proceedings of Machine Learning Research %D 2018 %E Jennifer Dy %E Andreas Krause %F pmlr-v80-mianjy18b %I PMLR %P 3540--3548 %U https://proceedings.mlr.press/v80/mianjy18b.html %V 80 %X Algorithmic approaches endow deep learning systems with implicit bias that helps them generalize even in over-parametrized settings. In this paper, we focus on understanding such a bias induced in learning through dropout, a popular technique to avoid overfitting in deep learning. For single hidden-layer linear neural networks, we show that dropout tends to make the norm of incoming/outgoing weight vectors of all the hidden nodes equal. In addition, we provide a complete characterization of the optimization landscape induced by dropout.
APA
Mianjy, P., Arora, R. & Vidal, R.. (2018). On the Implicit Bias of Dropout. Proceedings of the 35th International Conference on Machine Learning, in Proceedings of Machine Learning Research 80:3540-3548 Available from https://proceedings.mlr.press/v80/mianjy18b.html.

Related Material