Boundary between noise and information applied to filtering neural network weight matrices

Abstract

Deep neural networks have been successfully applied to a broad range of problems where overparametrization yields weight matrices which are partially random. A comparison of weight matrix singular vectors to the Porter-Thomas distribution suggests that there is a boundary between randomness and learned information in the singular value spectrum. Inspired by this finding, we introduce an algorithm for noise filtering, which both removes small singular values and reduces the magnitude of large singular values to counteract the effect of level repulsion between the noise and the information part of the spectrum. For networks trained in the presence of label noise, we indeed find that the generalization performance improves significantly due to noise filtering.

6 pages, 5 figures

References in corpus (5)

• NSFnets (Navier-Stokes Flow nets): Physics-informed neural networks for the incompressible Navier-Stokes equations
• Learning phase transitions by confusion
• Cleaning large correlation matrices: tools from random matrix theory
• Physics-informed generative neural network: an application to troposphere temperature prediction
• Random matrix analysis of deep neural network weight matrices

Cited by in corpus (1)

• Task complexity shapes internal representations and robustness in neural networks