Tensor-Fused Multi-View Graph Contrastive Learning

TL;DR

TensorMV-GCL introduces a tensor-fused multi-view graph contrastive learning framework that combines structural representations from augmented graphs with topological features derived from extended persistent homology. It employs a Tensor Transformation Layer to preserve multi-dimensional tensor structure and a stabilization strategy for EPH-based features, enabling robust joint learning through two contrastive objectives. The method achieves state-of-the-art performance on 9 of 11 graph-classification benchmarks across molecular, bioinformatic, and social datasets, with ablations confirming the importance of EPH, noise augmentation, and TTL. The approach offers a scalable, topology-aware alternative to existing GCL methods, and the authors provide code publicly.

Abstract

Graph contrastive learning (GCL) has emerged as a promising approach to enhance graph neural networks' (GNNs) ability to learn rich representations from unlabeled graph-structured data. However, current GCL models face challenges with computational demands and limited feature utilization, often relying only on basic graph properties like node degrees and edge attributes. This constrains their capacity to fully capture the complex topological characteristics of real-world phenomena represented by graphs. To address these limitations, we propose Tensor-Fused Multi-View Graph Contrastive Learning (TensorMV-GCL), a novel framework that integrates extended persistent homology (EPH) with GCL representations and facilitates multi-scale feature extraction. Our approach uniquely employs tensor aggregation and compression to fuse information from graph and topological features obtained from multiple augmented views of the same graph. By incorporating tensor concatenation and contraction modules, we reduce computational overhead by separating feature tensor aggregation and transformation. Furthermore, we enhance the quality of learned topological features and model robustness through noise-injected EPH. Experiments on molecular, bioinformatic, and social network datasets demonstrate TensorMV-GCL's superiority, outperforming 15 state-of-the-art methods in graph classification tasks across 9 out of 11 benchmarks while achieving comparable results on the remaining two. The code for this paper is publicly available at https://github.com/CS-SAIL/Tensor-MV-GCL.git.

Figures (1)

• Figure 1: In the pre-training stage (left), augmented graph views are processed through a shared GCN to generate node embeddings for Tensor-view Graph Contrastive Learning. In a parallel channel, a CNN is applied to extract features from EPH representations, with noise injected for robustness, then used for Stabilized Extended Persistent Images Contrastive Learning. Both channels adopt a TTL at the end to fuse the learned structural and topological features. In the downstream classification stage (right), a new graph is processed similarly, combining node embeddings with topological features extracted via the EPH-based CNN. The aggregated features pass through the TTL for final fusion, followed by an MLP for classification.