Monday, November 23, 9:00 a.m.
Johann Brehmer - New York University
How machine learning can help us get the most out of high-precision particle physics models
Abstract: Particle physics processes are usually modeled with a chain of complex simulators. While these allow us to generate high-quality simulated data, they face a challenging inverse problem when we want to infer theory parameters from observed data. I will explain why the relevant likelihood function cannot be evaluated, why that is a problem, discuss how particle physicists have circumvented it in the past, and introduce the framing of simulation-based inference. I will then show how we can use machine learning together with matrix-element information to solve it efficiently. Finally I will discuss how the same techniques can be useful in many other scientific fields from neuroscience to strong gravitational lensing.