New attack method TIRBA enhances GNN vulnerability to node injection

Researchers have developed a novel method called TIRBA (Target-aware Interaction-guided Reinforcement learning for Black-box node injection Attacks) to enhance the security of Graph Neural Networks (GNNs). TIRBA addresses the limitations of existing black-box node injection attacks by jointly optimizing the generation of malicious node features and the construction of edge connections within a heterogeneous action space. The approach utilizes a target-aware interaction encoder, a class-center guidance mechanism for efficient feature space exploration, and a topology difference-aware state value evaluation to stabilize training. Experimental results indicate that TIRBA significantly outperforms current state-of-the-art black-box node injection attack methods. AI

IMPACT This research highlights new vulnerabilities in GNNs, potentially driving further development in GNN security and defense mechanisms.

RANK_REASON Academic paper detailing a new method for attacking GNNs. [lever_c_demoted from research: ic=1 ai=1.0]

Read on arXiv cs.LG →

• Graph Neural Networks

• paper
• safety

AI-generated summary · Google Gemini · from 1 sources. How we write summaries →