Reproduce “Exploring the Limits of 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

Stanislav Fort, Jie Ren, Balaji Lakshminarayanan

The authors study detecting unfamiliar images with CIFAR-100 as known classes and CIFAR-10 as unknown classes. They adapt pretrained image representations to known classes, then score inputs using Mahalanobis distance or a classifier trained with labeled outliers. On benchmark, the fine-tuned hybrid model reaches 96.23% AUROC, versus 85.6% for a result; this does not establish detection of every unknown class.

Summary by alphaXiv.

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