Overview

Two-sample testing asks whether differences between two samples reflect a real change in their underlying distributions or ordinary sampling variation. Common approaches include kernel tests such as maximum mean discrepancy, classifier-based tests, nearest-neighbor and graph statistics, and permutation calibration; newer methods also use learned representations and generative-process dynamics. These tests support applications such as detecting data drift, comparing generative models, and finding changes in images, text, or networks. A central challenge is maintaining reliable false-alarm control while achieving power with high-dimensional, weakly different, or severely imbalanced samples.

Latest papers

CardsList

33 papers2 papers in the last four weeks, down 71% on the four weeks before. 0.0% of all new papers.

  1. Classical and quantum kernel fusion for two-sample testing

    Nov 26, 2025Yu Terada, Yugo Ogio, Ken Arai +2Two-Sample TestingMaximum Mean Discrepancy

  2. Efficient and Stable Multi-Dimensional Kolmogorov-Smirnov Distance

    Apr 15, 2025Peter Matthew Jacobs, Foad Namjoo, Jeff M. PhillipsTwo-Sample TestingDistribution Matching

  3. PQMass: Probabilistic Assessment of the Quality of Generative Models using Probability Mass Estimation

    Feb 6, 2024Pablo Lemos, Sammy Sharief, Esmeralda S. Whitammer +4Two-Sample TestingGenerative AI Evaluation