Explainability-Based Adversarial Attack on Graphs Through Edge Perturbation

TL;DR

This paper introduces an explainability-based adversarial attack method on graph neural networks that perturbs edges based on node importance, revealing vulnerabilities and emphasizing the impact of cross-class edge modifications.

Contribution

A novel explainability-driven approach for targeted edge perturbations in GNNs, enhancing understanding of attack impacts across multiple architectures and datasets.

Findings

Edge insertions between different classes have higher impact than edge removals within the same class.

The proposed method effectively identifies critical nodes for attack.

Cross-class edge perturbations significantly degrade GNN performance.

Abstract

Despite the success of graph neural networks (GNNs) in various domains, they exhibit susceptibility to adversarial attacks. Understanding these vulnerabilities is crucial for developing robust and secure applications. In this paper, we investigate the impact of test time adversarial attacks through edge perturbations which involve both edge insertions and deletions. A novel explainability-based method is proposed to identify important nodes in the graph and perform edge perturbation between these nodes. The proposed method is tested for node classification with three different architectures and datasets. The results suggest that introducing edges between nodes of different classes has higher impact as compared to removing edges among nodes within the same class.