Repository logo
 

Exact natural gradient in deep linear networks and application to the nonlinear case

Accepted version
Peer-reviewed

Loading...
Thumbnail Image

Type

Conference Object

Change log

Authors

Bernacchia, Alberto 
Lengyel, Máté 
Hennequin, Guillaume  ORCID logo  https://orcid.org/0000-0002-7296-6870

Abstract

Stochastic gradient descent (SGD) remains the method of choice for deep learning, despite the limitations arising for ill-behaved objective functions. In cases where it could be estimated, the natural gradient has proven very effective at mitigating the catastrophic effects of pathological curvature in the objective function, but little is known theoretically about its convergence properties, and it has yet to find a practical implementation that would scale to very deep and large networks. Here, we derive an exact expression for the natural gradient in deep linear networks, which exhibit pathological curvature similar to the nonlinear case. We provide for the first time an analytical solution for its convergence rate, showing that the loss decreases exponentially to the global minimum in parameter space. Our expression for the natural gradient is surprisingly simple, computationally tractable, and explains why some approximations proposed previously work well in practice. This opens new avenues for approximating the natural gradient in the nonlinear case, and we show in preliminary experiments that our online natural gradient descent outperforms SGD on MNIST autoencoding while sharing its computational simplicity.

Description

Keywords

Journal Title

32nd Conference on Neural Information Processing Systems (NIPS 2018), Montréal, Canada

Conference Name

Neural Information Processing Systems

Journal ISSN

1049-5258

Volume Title

Publisher

NIPS
Sponsorship
Wellcome Trust (202111/Z/16/Z)
Wellcome Trust (095621/Z/11/Z)
This work was supported by Wellcome Trust Seed Award 202111/Z/16/Z (G.H.) and Wellcome Trust Investigator Award 095621/Z/11/Z (A.B.,M.L.).