Top K Enhanced Reinforcement Learning Attacks on Heterogeneous Graph Node Classification

2024-08-04

arXiv.org

AI Summary

To address the insufficient robustness of heterogeneous graph node classification models under adversarial attacks, this paper proposes HeteroKRLAttack—the first black-box, target-oriented adversarial attack method specifically designed for heterogeneous graphs. HeteroKRLAttack innovatively integrates a Top-K pruning mechanism into a Proximal Policy Optimization (PPO) reinforcement learning framework, enabling efficient structural perturbation search within a constrained action space. It leverages meta-path modeling and gradient-free policy optimization while significantly compressing the high-dimensional discrete action space to enhance both attack efficiency and success rate. Extensive experiments on multiple heterogeneous graph benchmarks show that HeteroKRLAttack reduces the target model’s accuracy by an average of 32.7%. Ablation studies confirm that the Top-K module alone contributes a 19.4% improvement in attack success rate, underscoring its critical role in the overall framework.

Technology Category

• Machine Learning: Adversarial Learning & Robustness
• Computer Vision: Adversarial Attacks & Robustness
• Search and Optimization: Adversarial Search

Application Category

• Graph Algorithms and Modeling for the Web: Foundation models and LLMs for Web-related graphs
• User Modeling, Personalization and Recommendation: Attacks and countermeasures in recommendation systems
• Search and Retrieval-Augmented AI: Web learning to rank, online learning, and counterfactual learning for ranking

Abstract

Graph Neural Networks (GNNs) have attracted substantial interest due to their exceptional performance on graph-based data. However, their robustness, especially on heterogeneous graphs, remains underexplored, particularly against adversarial attacks. This paper proposes HeteroKRLAttack, a targeted evasion black-box attack method for heterogeneous graphs. By integrating reinforcement learning with a Top-K algorithm to reduce the action space, our method efficiently identifies effective attack strategies to disrupt node classification tasks. We validate the effectiveness of HeteroKRLAttack through experiments on multiple heterogeneous graph datasets, showing significant reductions in classification accuracy compared to baseline methods. An ablation study underscores the critical role of the Top-K algorithm in enhancing attack performance. Our findings highlight potential vulnerabilities in current models and provide guidance for future defense strategies against adversarial attacks on heterogeneous graphs.

Problem

• Attacking node classification on heterogeneous graphs
• Reducing action space using Top-K reinforcement learning
• Evaluating attack effectiveness on multiple graph datasets

Innovation

• Reinforcement learning with Top-K algorithm
• Targeted evasion black-box attack method
• Reduces action space for efficient attacks