Reproduce “Adversarial Purification by Consistency-aware Latent Space Optimization on Data Manifolds” 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, Jiahao Yang, Hui Luo, Jie Chen, Li Wang, Feng Liu, Bo Han, Mingkui Tan
CMAP purifies adversarial images for classification by optimizing latent vectors of a pretrained consistency model. It matches generated images to each input using pixel and structural similarity, regularizes the vectors toward the model’s prior, and aggregates predictions by voting. On CIFAR-10 under PGD+EOT, it reports 74.60% robust accuracy versus GNSP’s 55.87%; results remain attack-dependent.
Summary by alphaXiv.
Tip: change arxiv.org to autoarxiv.org in any arXiv abstract link to open the paper here.