Poisoning $\times$ Evasion: Symbiotic Adversarial Robustness for Graph Neural Networks

Ege Erdogan, Simon Geisler, Stephan Günnemann

Keywords: graph neural networks, adversarial robustness

TL;DR: Studies the robustness of graph neural networks by combining the evasion and poisoning threat models, showing that combining the threat models can lead to considerably stronger attacks.

Abstract: It is well-known that deep learning models are vulnerable w.r.t. small input perturbations. Such perturbed instances are called adversarial examples. Adversarial examples are commonly crafted to fool a model either at training time (poisoning) or test time (evasion). In this work, we study the symbiosis of poisoning and evasion. We show that combining both threat models can substantially improve the devastating efficacy of adversarial attacks. Specifically, we study the robustness of Graph Neural Networks (GNNs) under structure perturbations and devise a memory-efficient adaptive end-to-end attack for the novel threat model using first-order optimization.

Submission Number: 88