Randomized Schur Complement Views for Graph Contrastive Learning

TL;DR

This paper proposes a novel randomized augmentation method for Graph Contrastive Learning using Schur complements, which improves performance on node and graph classification tasks by providing unbiased graph views.

Contribution

It introduces a Schur complement-based topological augmentor for GCL, with theoretical justifications and empirical validation showing state-of-the-art results.

Findings

Consistently outperforms existing augmentation methods

Provides unbiased graph view approximations

Achieves state-of-the-art classification accuracy

Abstract

We introduce a randomized topological augmentor based on Schur complements for Graph Contrastive Learning (GCL). Given a graph laplacian matrix, the technique generates unbiased approximations of its Schur complements and treats the corresponding graphs as augmented views. We discuss the benefits of our approach, provide theoretical justifications and present connections with graph diffusion. Unlike previous efforts, we study the empirical effectiveness of the augmentor in a controlled fashion by varying the design choices for subsequent GCL phases, such as encoding and contrasting. Extensive experiments on node and graph classification benchmarks demonstrate that our technique consistently outperforms pre-defined and adaptive augmentation approaches to achieve state-of-the-art results.