Attacks on Node Attributes in Graph Neural Networks

TL;DR

This paper explores the vulnerability of graph neural networks to feature-based adversarial attacks on node attributes, demonstrating the effectiveness of decision time attacks over poisoning methods across multiple datasets.

Contribution

It introduces a focus on node attribute attacks in GNNs, contrasting with prior work that mainly targeted graph structure, and evaluates attack effectiveness using diverse datasets.

Findings

Decision time PGD attacks are more effective than poisoning attacks.

Node attribute attacks significantly compromise GNN performance.

Insights into vulnerabilities inform future defense strategies.

Abstract

Graphs are commonly used to model complex networks prevalent in modern social media and literacy applications. Our research investigates the vulnerability of these graphs through the application of feature based adversarial attacks, focusing on both decision time attacks and poisoning attacks. In contrast to state of the art models like Net Attack and Meta Attack, which target node attributes and graph structure, our study specifically targets node attributes. For our analysis, we utilized the text dataset Hellaswag and graph datasets Cora and CiteSeer, providing a diverse basis for evaluation. Our findings indicate that decision time attacks using Projected Gradient Descent (PGD) are more potent compared to poisoning attacks that employ Mean Node Embeddings and Graph Contrastive Learning strategies. This provides insights for graph data security, pinpointing where graph-based models are most vulnerable and thereby informing the development of stronger defense mechanisms against such attacks.