Abstract

The gradient noise of SGD is considered to play a central role in the observed strong generalization abilities of deep learning. While past studies confirm that the magnitude and covariance structure of gradient noise are critical for regularization, it remains unclear whether or not the class of noise distributions is important. In this work we provide negative results by showing that noises in classes different from the SGD noise can also effectively regularize gradient descent. Our finding is based on a novel observation on the structure of the SGD noise: it is the multiplication of the gradient matrix and a sampling noise that arises from the mini-batch sampling procedure. Moreover, the sampling noises unify two kinds of gradient regularizing noises that belong to the Gaussian class: the one using (scaled) Fisher as covariance and the one using the gradient covariance of SGD as covariance. Finally, thanks to the flexibility of choosing noise class, an algorithm is proposed to perform noisy gradient descent that generalizes well, the variant of which even benefits large batch SGD training without hurting generalization.

Meeting Name

37th International Conference on Machine Learning, ICML 2020 (2020: Jul. 13-18, Virtual)

Department(s)

Mathematics and Statistics

Comments

National Science Foundation, Grant JCKY2018204C004

International Standard Book Number (ISBN)

978-171382112-0

Document Type

Article - Conference proceedings

Document Version

Final Version

File Type

text

Language(s)

English

Rights

© 2020 The Authors, All rights reserved.

Publication Date

18 Jul 2020

Share

 
COinS