Graph Information Bottleneck

TL;DR

This paper introduces Graph Information Bottleneck (GIB), an information-theoretic approach to enhance the robustness of graph neural networks against adversarial attacks by balancing expressiveness and robustness.

Contribution

The paper proposes GIB, a novel regularization framework for GNNs that incorporates structural and feature information, improving adversarial robustness over existing models.

Findings

GIB models outperform state-of-the-art defenses against adversarial attacks.

GIB achieves up to 31% improvement under adversarial perturbations.

Structural and feature regularization enhances GNN robustness.

Abstract

Representation learning of graph-structured data is challenging because both graph structure and node features carry important information. Graph Neural Networks (GNNs) provide an expressive way to fuse information from network structure and node features. However, GNNs are prone to adversarial attacks. Here we introduce Graph Information Bottleneck (GIB), an information-theoretic principle that optimally balances expressiveness and robustness of the learned representation of graph-structured data. Inheriting from the general Information Bottleneck (IB), GIB aims to learn the minimal sufficient representation for a given task by maximizing the mutual information between the representation and the target, and simultaneously constraining the mutual information between the representation and the input data. Different from the general IB, GIB regularizes the structural as well as the feature information. We design two sampling algorithms for structural regularization and instantiate the GIB principle with two new models: GIB-Cat and GIB-Bern, and demonstrate the benefits by evaluating the resilience to adversarial attacks. We show that our proposed models are more robust than state-of-the-art graph defense models. GIB-based models empirically achieve up to 31% improvement with adversarial perturbation of the graph structure as well as node features.