2022-11-03 Stat Cafe - Jian Yan

• Time: Monday 11/03 from 2:30 PM to 3:30 PM
• Location: BLOC 503
• Snacks and drinks will be provided
• Gallery

Topic

The history and some recent advances of distance-based and RKHS-based two-sample tests and some other tests

Abstract

Two-sample testing is a classical and fundamental problem in statistics. Modern nonparametric tests have been developed using energy distance (ED) and maximum mean discrepancy (MMD), which have been shown to be equivalent. In this talk, I’ll first give a brief overview of the history of these two metrics and how they can be applied to the two-sample testing. Then I’ll highlight some recent advances in understanding these metrics and associated two-sample tests in high dimension, which are quite interesting and surprising. Later, I’ll also introduce other tests based on ED and MMD, including independence test, conditional mean independence test, two-sample conditional distribution test, and conditional independence test. The relationships between different tests will be discussed.