Adversarial Examples for Graph Data: Deep Insights into Attack and Defense

Huijun Wu, Chen Wang, Yuriy Tyshetskiy, Andrew Docherty, Kai Lü, Liming Zhu

2019 · 394 citations · 19 references

DOIFull text

Open access

Concepts

TL;DR

Graph deep learning models such as GCN achieve state‑of‑the‑art performance but, like other deep learning models, are vulnerable to adversarial attacks, and the discrete nature of graph connections and features creates unique challenges and opportunities for such attacks. This work proposes both an adversarial attack and a defense strategy for graph data. The attack and defense rely on integrated‑gradient analysis to accurately assess the impact of modifying discrete features or edges, enabling targeted perturbations and detection/recovery of adversarial changes. Targeted attacks produce statistically distinguishable graphs, and experiments on multiple datasets demonstrate the effectiveness of the proposed attack and defense methods.

Abstract

Graph deep learning models, such as graph convolutional networks (GCN) achieve state-of-the-art performance for tasks on graph data. However, similar to other deep learning models, graph deep learning models are susceptible to adversarial attacks. However, compared with non-graph data the discrete nature of the graph connections and features provide unique challenges and opportunities for adversarial attacks and defenses. In this paper, we propose techniques for both an adversarial attack and a defense against adversarial attacks. Firstly, we show that the problem of discrete graph connections and the discrete features of common datasets can be handled by using the integrated gradient technique that accurately determines the effect of changing selected features or edges while still benefiting from parallel computations. In addition, we show that an adversarially manipulated graph using a targeted attack statistically differs from un-manipulated graphs. Based on this observation, we propose a defense approach which can detect and recover a potential adversarial perturbation. Our experiments on a number of datasets show the effectiveness of the proposed techniques.

References

19