Variational Graph Contrastive Learning

TL;DR

This paper introduces SGEC, a novel self-supervised graph representation learning method that uses subgraph Gaussian embeddings and optimal transport distances to improve contrastive learning of graph data.

Contribution

The paper proposes a new subgraph Gaussian embedding module and employs optimal transport distances to enhance contrastive learning in graph representation learning.

Findings

SGEC outperforms existing methods on multiple benchmarks.

The use of optimal transport distances improves robustness.

Distribution control of contrastive pairs is crucial for performance.

Abstract

Graph representation learning (GRL) is a fundamental task in machine learning, aiming to encode high-dimensional graph-structured data into low-dimensional vectors. Self-supervised learning (SSL) methods are widely used in GRL because they can avoid expensive human annotation. In this work, we propose a novel Subgraph Gaussian Embedding Contrast (SGEC) method. Our approach introduces a subgraph Gaussian embedding module, which adaptively maps subgraphs to a structured Gaussian space, ensuring the preservation of graph characteristics while controlling the distribution of generated subgraphs. We employ optimal transport distances, including Wasserstein and Gromov-Wasserstein distances, to effectively measure the similarity between subgraphs, enhancing the robustness of the contrastive learning process. Extensive experiments across multiple benchmarks demonstrate that SGEC outperforms or presents competitive performance against state-of-the-art approaches. Our findings provide insights into the design of SSL methods for GRL, emphasizing the importance of the distribution of the generated contrastive pairs.