Reproduce “A Witness Two-Sample Test” 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

Jonas M. Kübler, Wittawat Jitkrittum, Bernhard Schölkopf, Krikamol Muandet

The authors introduce WiTS, a two-sample test that learns a one-dimensional witness from training data and compares its average on test data. The KFDA witness weights differences in kernel-space means by an inverse-covariance-like adjustment. On two benchmarks, both witness tests show high rejection rates as sample size grows, and MMD-D-witness improves on MMD-D; performance depends on the split and tuning choices.

Summary by alphaXiv.

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