Reproduce “Watermarking for Out-of-distribution Detection” 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

Qizhou Wang, Feng Liu, Yonggang Zhang, Jing Zhang, Chen Gong, Tongliang Liu, Bo Han

The method learns a watermark for a fixed classifier, adding it to inputs without changing classifier weights. It optimizes the pattern using ID examples and Gaussian-noise negatives, then applies a score. For a WideResNet-40-2 trained on CIFAR-10, free-energy detection across five OOD datasets reduces FPR95 from 37.67% to 23.69% and raises AUROC from 90.56% to 95.21%, not establishing universal performance.

Summary by alphaXiv.

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