Homophily modulates double descent generalization in graph convolution networks

TL;DR

This paper uses statistical physics and random matrix theory to analyze how homophily influences double descent phenomena in graph convolutional networks, revealing insights into their generalization behavior on various data types.

Contribution

It provides a theoretical characterization of GNN generalization, especially regarding double descent and the effects of homophily versus heterophily, supported by analytical and empirical results.

Findings

Double descent occurs in GNNs and is influenced by data homophily.

Risk depends on graph noise, feature noise, and training labels.

Analytic insights improve GNN performance on heterophilic datasets.

Abstract

Graph neural networks (GNNs) excel in modeling relational data such as biological, social, and transportation networks, but the underpinnings of their success are not well understood. Traditional complexity measures from statistical learning theory fail to account for observed phenomena like the double descent or the impact of relational semantics on generalization error. Motivated by experimental observations of ``transductive'' double descent in key networks and datasets, we use analytical tools from statistical physics and random matrix theory to precisely characterize generalization in simple graph convolution networks on the contextual stochastic block model. Our results illuminate the nuances of learning on homophilic versus heterophilic data and predict double descent whose existence in GNNs has been questioned by recent work. We show how risk is shaped by the interplay between the graph noise, feature noise, and the number of training labels. Our findings apply beyond stylized models, capturing qualitative trends in real-world GNNs and datasets. As a case in point, we use our analytic insights to improve performance of state-of-the-art graph convolution networks on heterophilic datasets.