Subgroup Generalization and Fairness of Graph Neural Networks

TL;DR

This paper introduces a PAC-Bayesian framework to analyze the generalization and fairness of graph neural networks, especially for non-IID node data, highlighting the importance of training node selection for equitable performance.

Contribution

It presents a novel theoretical analysis of GNN generalization and fairness using PAC-Bayesian methods in non-IID settings, emphasizing subgroup performance disparities.

Findings

Performance on subgroups depends on the distance to training data.

Training node selection significantly impacts fairness.

Experimental results support the theoretical analysis.

Abstract

Despite enormous successful applications of graph neural networks (GNNs), theoretical understanding of their generalization ability, especially for node-level tasks where data are not independent and identically-distributed (IID), has been sparse. The theoretical investigation of the generalization performance is beneficial for understanding fundamental issues (such as fairness) of GNN models and designing better learning methods. In this paper, we present a novel PAC-Bayesian analysis for GNNs under a non-IID semi-supervised learning setup. Moreover, we analyze the generalization performances on different subgroups of unlabeled nodes, which allows us to further study an accuracy-(dis)parity-style (un)fairness of GNNs from a theoretical perspective. Under reasonable assumptions, we demonstrate that the distance between a test subgroup and the training set can be a key factor affecting the GNN performance on that subgroup, which calls special attention to the training node selection for fair learning. Experiments across multiple GNN models and datasets support our theoretical results.