High-Probability Mixing Time Bound for Gibbs Sampling from Log-Smooth Strongly Log-Concave Distributions

Explore a technical seminar presentation where Dr. Neha Spenta Wadia from the Simons Foundation discusses high-probability mixing time bounds for Gibbs sampling from log-smooth strongly log-concave distributions. Delve into the mathematical foundations of Markov Chain Monte Carlo (MCMC) sampling techniques, focusing on the Gibbs sampler's application to probability distributions in n-dimensional space. Learn how recent developments, building upon work by Aditi Laddha and Santosh Vempala, establish polynomial mixing time bounds for uniform distributions on convex bodies. Understand the proof methodology involving conductance arguments and a novel high-probability L0 isoperimetric inequality for convex body subsets. Part of the Stochastic Systems for Anomalous Diffusion series at the Isaac Newton Institute, this hour-long presentation advances our understanding of non-asymptotic convergence guarantees in high-dimensional sampling.