Workshop 1 on Sequential Hypothesis Testing

Description
Data, in many cases, arrive continuously and results are examined as they accrue. But classical tests are valid only at a single sample size fixed in advance, and examining results repeatedly increases the false-positive rate. This workshop covers recent work on e-values and safe anytime-valid inference, which measure evidence as the accumulated wealth of a bet against the null hypothesis and stay valid however often the data are examined. For example: composite hypotheses such as “this strategy has no edge” or “this model remains calibrated.” We connect the e-value construction to Wald’s sequential probability ratio test, Kelly betting and Cover’s universal portfolio, and close with a discussion of some applications.

This 3-hour workshop, which is a part of the 2026 AI Applications in Industry Conference, is organized in the form of a tutorial. It aims at introducing practitioners (both from academia and industry) to statistical and econometric techniques that have broad applicability which researchers are working on. During the workshop participants will exchange ideas on limitations of current techniques, go into the details of the novel paradigm that addresses some of these limitations, and discuss areas where the novel paradigm may be applied. Seats for this workshop are very limited. Only 15 spots are available.

Instructor: Patrick Flaherty, PhD (University of Massachusetts Amherst)

Date: Thursday, September 24, 2026
Time: 10:00am-1:00pm
Venue: Third Floor, AV3 Meeting Room, Federal Reserve Plaza, Boston, MA
Price: $50.00