E-values and e-processes are emerging in many hypothesis testing situations, including the different paradigms for multiple testing (FDR control, closed testing, etc.), and they turn out to have interesting properties. One of the key merits of the e-process, is the possibility to analyse data sequentially in time, and in this talk I will discuss how to use e-processes for multiple testing, eventually leading to a framework where the researcher can add both hypotheses and data at will while retaining strong error guarantees.
Period
17 Dec 2024
Event title
IMS International Conference on Statistics and Data Science, ICSDS 2024