Reproduce “Detecting Adversarial Data by Probing Multiple Perturbations Using Expected Perturbation Score” on OpenResearch
Turn your favorite coding agents into research agents that review literature, develop hypotheses, and run experiments.
Universal app for Apple silicon and Intel. Requires macOS 11 or later.
About this paper
Shuhai Zhang, Feng Liu, Jiahao Yang, Yifan Yang, Changsheng Li, Bo Han, Mingkui Tan
EPS-AD detects adversarial examples by averaging diffusion-model scores across noise levels and comparing the expected perturbation score vectors for a test image with vectors from natural reference images using maximum mean discrepancy. On ImageNet with ResNet-50, it reports AUROC 1.0000 for FGSM-ℓ₂ at ε=4/255, versus 0.8191 for an EPS-norm baseline in a 500-versus-500 evaluation; this result applies to that setup.
Summary by alphaXiv.
Tip: change arxiv.org to autoarxiv.org in any arXiv abstract link to open the paper here.