Towards Robust Graph Contrastive Learning

TL;DR

This paper proposes a new graph contrastive learning method that enhances adversarial robustness by using adversarial transformations, including edge removal and insertion, showing promising preliminary results.

Contribution

Introduces a novel approach to improve robustness in graph contrastive learning through adversarial transformations, including edge insertion and removal.

Findings

Preliminary experiments show promising robustness improvements.

The method effectively incorporates adversarial transformations.

Results suggest potential for robustness as an auxiliary task.

Abstract

We study the problem of adversarially robust self-supervised learning on graphs. In the contrastive learning framework, we introduce a new method that increases the adversarial robustness of the learned representations through i) adversarial transformations and ii) transformations that not only remove but also insert edges. We evaluate the learned representations in a preliminary set of experiments, obtaining promising results. We believe this work takes an important step towards incorporating robustness as a viable auxiliary task in graph contrastive learning.