Reproduce “STROKE: Simple Sparse Structural Occlusions for Black-Box Adversarial Attacks” 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

Ankit Gupta, Akshat Pandya, Paul Lin, Christoph Adami, Emily Dolson

STROKE is a score-based black-box attack that uses oriented rectangular occlusions and greedy addition and refinement to mislead image classifiers. In a matched untargeted ImageNet ResNet-50 test at 2% coverage and a 10,000-round limit, it achieved 98.4% success, with 796 average queries and 0.22 seconds per image, compared with IMPACT. Targeted success was lower than IMPACT’s in a reported comparison.

Summary by alphaXiv.

Tip: change arxiv.org to autoarxiv.org in any arXiv abstract link to open the paper here.