Robustness evaluation for deep neural networks via mutation decision boundaries analysis

Renhao Lin, Qinglei Zhou, Bin Wu, Xiaofei Nan

Abstract: Highlights•Proposing a local robustness evaluation method as a pre-analysis before the verification.•Searching for unstable points for priority verification and adversarial examples generation.•Designing a prioritization-based verification framework to improve verification efficiency.•Evaluating on verficication tools to demonstrate the effectiveness of our solution.