Assistant Professor, University of Toronto
Talk Title
Sampling with Langevin Monte Carlo: Recent Progress
Abstract
This talk will focus on the recent progress on the theory of sampling from a target distribution $e^{-f}$ using the Langevin Monte Carlo (LMC) algorithm. In particular, we investigate the sufficient number of steps to reach the $\epsilon$-neighborhood of a $d$-dimensional target distribution as a function of tail-growth and smoothness of the potential $f$. The results will cover the first convergence guarantees for LMC under several functional inequalities such as the weak-Poincar\'e, Poincar\'e and log-Sobolev settings.