Agentic data science with the Data Agent Kit
cloud.google.com·8h ago
AI Research
arXiv CS.AI·Sep 9·5 min read
Protocol effects on feature-based hardware-Trojan detection across Trust-Hub families
TL;DR
Research evaluates how different test boundary settings affect the performance of hardware-Trojan detection models across Trust-Hub families. Results show that withholding entire netlists or host families significantly reduces detection accuracy.
✦ Why It Matters
Ensure your hardware-Trojan detection benchmarks include family-aware holdouts to avoid inflated performance metrics.
Key Takeaways
- Use family-aware holdouts in Trojan detection benchmarks.
- Be cautious of inflated performance metrics from sibling variants.
- Evaluate detection models with varied test boundaries.
- Report both pooled and family-specific results for clarity.
How It Works
The study measures the impact of test boundary settings on detection accuracy by comparing models trained on pooled data versus those tested on withheld netlists or families. This approach reveals how prior exposure to host logic can influence model performance.