Student seminars

#StudentSeminar: Sampling from complex probability distributions: from Monte Carlo methods to machine-learning normalizing flows

by David Albandea (IFIC, CSIC-UV)

Europe/Madrid
1001-Primera-1-1-1 - Paterna. Seminario (Universe)

1001-Primera-1-1-1 - Paterna. Seminario

Universe

60
Description

Abstract:
Being able to obtain samples from a probability distribution is key in many scientific applications, such as statistical inference and the simulation of statistical systems. In particular, Quantum Chromodynamics (QCD) in discretized spacetime can be treated as a statistical system, and its study boils down to obtaining samples from a very complex and high-dimensional probability distribution. In this talk I will try to do a friendly exploration of the foundations of sampling methods, introducing Monte Carlo techniques such as importance sampling and Markov Chain Monte Carlo (MCMC) algorithms. This will pave the way to talk about a modern machine learning sampling technique known as normalizing flows, and we will discuss its possible applications for lattice QCD simulations.

 

(Theoretical physics)