Enhancing Size Generalization in Graph Neural Networks through Disentangled Representation Learning

TL;DR

This paper introduces DISGEN, a framework that improves the ability of graph neural networks to generalize across different graph sizes by disentangling size information from representations, leading to better performance.

Contribution

DISGEN is a novel, model-agnostic framework that effectively removes size information from graph representations, enhancing size generalization in GNNs.

Findings

DISGEN outperforms state-of-the-art models by up to 6% on real-world datasets.

The framework employs size- and task-invariant augmentations and a decoupling loss.

Theoretical guarantees support the effectiveness of the disentanglement approach.

Abstract

Although most graph neural networks (GNNs) can operate on graphs of any size, their classification performance often declines on graphs larger than those encountered during training. Existing methods insufficiently address the removal of size information from graph representations, resulting in sub-optimal performance and reliance on backbone models. In response, we propose DISGEN, a novel and model-agnostic framework designed to disentangle size factors from graph representations. DISGEN employs size- and task-invariant augmentations and introduces a decoupling loss that minimizes shared information in hidden representations, with theoretical guarantees for its effectiveness. Our empirical results show that DISGEN outperforms the state-of-the-art models by up to 6% on real-world datasets, underscoring its effectiveness in enhancing the size generalizability of GNNs. Our codes are available at: https://github.com/GraphmindDartmouth/DISGEN.