How does Weight Correlation Affect the Generalisation Ability of Deep Neural Networks



Jin, Gaojie, Yi, Xinping ORCID: 0000-0001-5163-2364, Zhang, Liang, Zhang, Lijun, Schewe, Sven ORCID: 0000-0002-9093-9518 and Huang, Xiaowei ORCID: 0000-0001-6267-0366
(2020) How does Weight Correlation Affect the Generalisation Ability of Deep Neural Networks. .

[img] Text
2010.05983v2.pdf - Author Accepted Manuscript

Download (2MB) | Preview

Abstract

This paper studies the novel concept of weight correlation in deep neural networks and discusses its impact on the networks' generalisation ability. For fully-connected layers, the weight correlation is defined as the average cosine similarity between weight vectors of neurons, and for convolutional layers, the weight correlation is defined as the cosine similarity between filter matrices. Theoretically, we show that, weight correlation can, and should, be incorporated into the PAC Bayesian framework for the generalisation of neural networks, and the resulting generalisation bound is monotonic with respect to the weight correlation. We formulate a new complexity measure, which lifts the PAC Bayes measure with weight correlation, and experimentally confirm that it is able to rank the generalisation errors of a set of networks more precisely than existing measures. More importantly, we develop a new regulariser for training, and provide extensive experiments that show that the generalisation error can be greatly reduced with our novel approach.

Item Type: Conference or Workshop Item (Unspecified)
Additional Information: Accpeted by NeurIPS 2020 conference
Uncontrolled Keywords: cs.LG, cs.LG, cs.AI
Depositing User: Symplectic Admin
Date Deposited: 19 Oct 2020 09:30
Last Modified: 18 Jan 2023 23:28
Related URLs:
URI: https://livrepository.liverpool.ac.uk/id/eprint/3104426