A0378
Title: A cross-validation approach for distribution-free two-sample testing with high-dimensional data
Authors: Shunan Yao - Hong Kong Baptist University (Hong Kong) [presenting]
Abstract: Two-sample testing is a cornerstone of modern statistics. In the realm of high-dimensional two-sample testing, traditional methods often rely on distributional assumptions, e.g., Hotelling's T-test, or use a subset of the data for dimension reduction, thereby not utilizing the full dataset for the actual test. A novel cross-validation style approach is introduced to two-sample testing that is computationally tractable and capable of using the full dataset for testing. Specifically, (a) it is shown that the method achieves super-uniformity in finite samples under the null hypothesis, where both samples originate from an identical distribution, (b) establish the asymptotic properties of the test statistic, and (c) the performance is demonstrated of the method through numerical applications.
