Gradient Leakage Attacks Pose Security Risks to GNNs in Circuit Design

Researchers have identified significant security and privacy risks associated with gradient leakage attacks (GLAs) on Graph Neural Networks (GNNs) used in circuit design and hardware security. These attacks can expose sensitive information like gate types and hardware Trojan properties, potentially aiding adversaries in logic locking analysis or evading detection. While some defense techniques show limited effectiveness and can degrade model performance, architectural choices like attention mechanisms (GAT) can worsen leakage, whereas injective aggregation (GIN) offers better resilience. AI

IMPACT Highlights potential security vulnerabilities in AI models used for critical infrastructure, necessitating more robust privacy-preserving techniques.

RANK_REASON Academic paper detailing a new attack vector and analysis of defenses on GNNs. [lever_c_demoted from research: ic=1 ai=1.0]

Read on arXiv cs.LG →

• adversarial training
• differential privacy
• gradient clipping
• Gradient Leakage Attacks
• GAT
• Graph Neural Networks
• GraphSAGE
• ISCAS'85
• secure aggregation
• EPFL
• TrustHub

• paper
• safety

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