Calibrating and Improving Graph Contrastive Learning

TL;DR

This paper introduces Contrast-Reg, a regularization method for graph contrastive learning that improves downstream task performance by aligning contrastive loss reduction with true task objectives, supported by theoretical and empirical evidence.

Contribution

The paper proposes a novel calibration-based regularizer, Contrast-Reg, for graph contrastive learning that enhances generalization and downstream task performance.

Findings

Contrast-Reg improves GNN generalizability across tasks

Enhances performance of various contrastive algorithms

Theoretically and empirically validated effectiveness

Abstract

Graph contrastive learning algorithms have demonstrated remarkable success in various applications such as node classification, link prediction, and graph clustering. However, in unsupervised graph contrastive learning, some contrastive pairs may contradict the truths in downstream tasks and thus the decrease of losses on these pairs undesirably harms the performance in the downstream tasks. To assess the discrepancy between the prediction and the ground-truth in the downstream tasks for these contrastive pairs, we adapt the expected calibration error (ECE) to graph contrastive learning. The analysis of ECE motivates us to propose a novel regularization method, Contrast-Reg, to ensure that decreasing the contrastive loss leads to better performance in the downstream tasks. As a plug-in regularizer, Contrast-Reg effectively improves the performance of existing graph contrastive learning algorithms. We provide both theoretical and empirical results to demonstrate the effectiveness of Contrast-Reg in enhancing the generalizability of the Graph Neural Network(GNN) model and improving the performance of graph contrastive algorithms with different similarity definitions and encoder backbones across various downstream tasks.