An introduction to minimax hypothesis testing

Speaker: Lucas Kania, PhD student at Carnegie Mellon University (CMU)

Abstract:
Historically, research in hypothesis testing has focused on finding tests with asymptotically tractable limits and studying the assumptions under which those limits hold. Albeit such perspective is successful in the low-dimensional and big-data setting, it excludes many cases of interest: high-dimensional and small-data problems where tailored tests can have high power despite having unusual asymptotic distributions. In this talk, we will introduce the audience to the non-asymptotic minimax framework and argue that it can aid us in finding powerful and practical tests in all settings of interest. We will explore the framework in several problems with discrete and continuous data. Finally, we will discuss open problems related to the theory and practice of hypothesis testing.

Host: Prof. Ernst Wit