From Bi-Level to One-Level: A Framework for Structural Attacks to Graph Anomaly Detection

TL;DR

This paper introduces BinarizedAttack, a novel method that simplifies and enhances structural poisoning attacks on graph anomaly detection systems by transforming complex bi-level problems into more manageable one-level optimization problems.

Contribution

It proposes a new framework converting bi-level optimization into one-level, utilizing gradient information for effective structural attacks on GAD systems.

Findings

BinarizedAttack outperforms existing attack methods.

Structural vulnerabilities are effectively exploited.

The approach is applicable to both supervised and unsupervised GAD systems.

Abstract

The success of graph neural networks stimulates the prosperity of graph mining and the corresponding downstream tasks including graph anomaly detection (GAD). However, it has been explored that those graph mining methods are vulnerable to structural manipulations on relational data. That is, the attacker can maliciously perturb the graph structures to assist the target nodes to evade anomaly detection. In this paper, we explore the structural vulnerability of two typical GAD systems: unsupervised FeXtra-based GAD and supervised GCN-based GAD. Specifically, structural poisoning attacks against GAD are formulated as complex bi-level optimization problems. Our first major contribution is then to transform the bi-level problem into one-level leveraging different regression methods. Furthermore, we propose a new way of utilizing gradient information to optimize the one-level optimization problem in the discrete domain. Comprehensive experiments demonstrate the effectiveness of our proposed attack algorithm BinarizedAttack.