Adversarial Contrastive Graph Augmentation with Counterfactual Regularization

Tao Long, Liang Zhang

Keywords: Node Representation Learning, Contrastive Learning, Node Classification

TL;DR: A novel model-based adversarial contrastive graph augmentation (ACGA) method for generating both positive samples with minimal sufficient information and hard negative graph samples automatically.

Abstract:

With the development of graph representation learning, self-supervised graph contrastive learning (GCL) has become one of the most important techniques. In GCL, a set of positive and negative samples are generated by data augmentation. However, the majority of existing methods rely on empirical rule-based graph augmentations, which might lead to failures in learning useful graph patterns. To address this issue, we propose a novel model-based adversarial contrastive graph augmentation (ACGA) method for generating both positive samples with minimal sufficient information and hard negative graph samples automatically. We also provide a theoretical framework to analyze the positive and negative graph augmenting process in self-supervised GCL. We evaluate our ACGA via extensive experiments on five benchmark datasets. The experimental results show that ACGA outperforms the state-of-the-art baselines.

Primary Area: unsupervised, self-supervised, semi-supervised, and supervised representation learning

Code Of Ethics: I acknowledge that I and all co-authors of this work have read and commit to adhering to the ICLR Code of Ethics.

Submission Guidelines: I certify that this submission complies with the submission instructions as described on https://iclr.cc/Conferences/2024/AuthorGuide.

Anonymous Url: I certify that there is no URL (e.g., github page) that could be used to find authors' identity.

No Acknowledgement Section: I certify that there is no acknowledgement section in this submission for double blind review.

Submission Number: 637