GPS: Graph Contrastive Learning via Multi-scale Augmented Views from Adversarial Pooling

TL;DR

GPS introduces a novel graph contrastive learning method that automatically generates multi-scale augmented views using adversarial pooling, improving robustness and performance without relying on predefined augmentation strategies.

Contribution

The paper proposes a new graph contrastive learning approach leveraging adversarially trained graph pooling to automatically create challenging multi-scale views, enhancing representation learning.

Findings

Outperforms existing methods on twelve datasets.

Demonstrates improved robustness and transferability.

Effectively generates challenging augmented views.

Abstract

Self-supervised graph representation learning has recently shown considerable promise in a range of fields, including bioinformatics and social networks. A large number of graph contrastive learning approaches have shown promising performance for representation learning on graphs, which train models by maximizing agreement between original graphs and their augmented views (i.e., positive views). Unfortunately, these methods usually involve pre-defined augmentation strategies based on the knowledge of human experts. Moreover, these strategies may fail to generate challenging positive views to provide sufficient supervision signals. In this paper, we present a novel approach named Graph Pooling ContraSt (GPS) to address these issues. Motivated by the fact that graph pooling can adaptively coarsen the graph with the removal of redundancy, we rethink graph pooling and leverage it to automatically generate multi-scale positive views with varying emphasis on providing challenging positives and preserving semantics, i.e., strongly-augmented view and weakly-augmented view. Then, we incorporate both views into a joint contrastive learning framework with similarity learning and consistency learning, where our pooling module is adversarially trained with respect to the encoder for adversarial robustness. Experiments on twelve datasets on both graph classification and transfer learning tasks verify the superiority of the proposed method over its counterparts.