Reproduce “AutoML 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, Vincent Stimper, Simon Buchholz, Krikamol Muandet, Bernhard Schölkopf
The authors present an AutoML two-sample test that learns a real-valued witness to distinguish two datasets, then tests its mean-score difference on held-out examples. A weighted squared-loss objective is theoretically optimal for maximizing asymptotic test power when the distributions’ supports overlap. Across simulated MNIST and CIFAR10 shifts, it is competitive at larger sample sizes, but not the very smallest.
Summary by alphaXiv.
Tip: change arxiv.org to autoarxiv.org in any arXiv abstract link to open the paper here.