Over-smoothing Effect of Graph Convolutional Networks

TL;DR

This paper analyzes the over-smoothing problem in Graph Convolutional Networks, providing an upper bound for its occurrence and explaining the effectiveness of algorithms designed to mitigate it.

Contribution

It offers a comprehensive analysis of over-smoothing in GCNs, introducing an upper bound and explaining why certain algorithms can alleviate this issue.

Findings

Established an upper bound for over-smoothing in GCNs

Explained the mechanisms behind over-smoothing

Validated the effectiveness of algorithms that mitigate over-smoothing

Abstract

Over-smoothing is a severe problem which limits the depth of Graph Convolutional Networks. This article gives a comprehensive analysis of the mechanism behind Graph Convolutional Networks and the over-smoothing effect. The article proposes an upper bound for the occurrence of over-smoothing, which offers insight into the key factors behind over-smoothing. The results presented in this article successfully explain the feasibility of several algorithms that alleviate over-smoothing.