Reproduce “LOTTERY: Learning from Reference-Only Samples in Two-Sample Testing under Size Asymmetry” 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
Xunye Tian, Zhijian Zhou, Liuhua Peng, Feng Liu
LOTTERY tests whether small query batches differ from a reference sample by learning compatibility scores from reference data alone. It splits reference data for learning, calibration, and held-out permutation testing, leaving queries for testing. On adversarial CIFAR-10 embeddings with 2 queries and 100 references, power is 0.810 ± 0.011 versus MMD-M’s 0.146 ± 0.012; this result applies to that setting.
Summary by alphaXiv.
Tip: change arxiv.org to autoarxiv.org in any arXiv abstract link to open the paper here.