PHYSTAT Seminar: Likelihood free Inference

by Harry Prosper (Florida State University (US))

Wednesday 27 Sept 2023, 16:00 → 17:30 Europe/Zurich

Abstract:High-fidelity simulators when coupled with machine learning make it possible to infer the parameters of a theoretical model directly from real and simulated data without explicit use of the likelihood function. This is of particular interest when the likelihood is intractable as is often the case for unbinned analyses. Recently, Prof. Ann Lee and her group introduced the likelihood-free frequenist (LF2I) inference approach, which features frequentist guarantees for finite samples. In this talk, I introduce a simple modification of LF2I that has some computational advantages. After a brief, pedagogical, review of the connection between hypothesis testing and confidence sets, I illustrate the utility of the modified algorithm by applying it to three simple examples: the first is from cosmology, the socond is from high-energy physics, while the third is from epidemiology.

Harrison Prosper is an experimental particle physicist and the Kirby W. Kemper Endowed Professor of Physics at Florida State University. He has a special interest in Bayesian methods and advanced data analysis techniques including machine learning. He has performed experiments at DESY, ILL, Fermilab, and CERN, an he is an active member of the CMS Coallaboration at the LHC.

 

 

 

 

GMT20230927-140151_Recording_1760x900.mp4

PHYSTAT_09_27_2023.pdf