Hardware Trojan detection inflated by sibling variants, study finds

A new research paper explores the impact of protocol effects on hardware-Trojan detection within the Trust-Hub families. The study found that when sibling benchmark variants are included in both training and testing datasets, detectors can achieve inflated performance metrics by leveraging previously seen host logic. This effect was measured across different machine learning models, including random forest, XGBoost, and logistic regression, with performance significantly dropping when host families were held out from the test set. The researchers recommend that benchmarks with multiple variants of a host circuit should report family-aware holdouts alongside pooled scores to provide a more accurate assessment of detection capabilities. AI

IMPACT Highlights potential overestimation of model performance in hardware security benchmarks, suggesting a need for more rigorous evaluation protocols.

RANK_REASON Academic paper on AI/ML research methodology. [lever_c_demoted from research: ic=1 ai=1.0]

Read on arXiv cs.AI →

• Institute of Software, Chinese Academy of Sciences
• logistic regression model
• random forest
• Trust-Hub
• XGBoost

• paper
• safety

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