Implicit Bias of Gradient Descent on Linear Convolutional Networks

We show that gradient descent on full-width linear convolutional networks of depth \(L\) converges to a linear predictor related to the \(\ell_{2/L}\) bridge penalty in the frequency domain. This is in contrast to linearly fully connected networks, where gradient descent converges to the hard margin...

Full description

Saved in:
Bibliographic Details
Published in:arXiv.org 2019-01
Main Authors: Gunasekar, Suriya, Lee, Jason, Soudry, Daniel, Srebro, Nathan
Format: Article
Language:English
Subjects:
Artificial neural networks
Convergence
Support vector machines
Online Access:Get full text
Tags: Add Tag
No Tags, Be the first to tag this record!
Description
Summary:We show that gradient descent on full-width linear convolutional networks of depth \(L\) converges to a linear predictor related to the \(\ell_{2/L}\) bridge penalty in the frequency domain. This is in contrast to linearly fully connected networks, where gradient descent converges to the hard margin linear support vector machine solution, regardless of depth.
ISSN:2331-8422