Reproduce “Robustness Verification against Patch Attacks via Group-Lasso Semidefinite Relaxation” 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
Yukina Nakagawa, Takami Sato, Ken Kobayashi, Kazuhide Nakata
The paper develops a verifier for localized image patch attacks by grouping pixels by possible patch location and replacing discrete group selection with a continuous Group-Lasso relaxation. The relaxation contains the original patch threat, so a certificate applies to that threat, though failure to certify is inconclusive. In adversarially trained FCNNa on MNIST at ε = 0.10, it certified 91% of tested samples.
Summary by alphaXiv.
Tip: change arxiv.org to autoarxiv.org in any arXiv abstract link to open the paper here.