Statistics Colloquium: Yves Atchade (Boston University)

Date: 

Monday, October 5, 2020, 10:30am to 11:30am

Location: 

Zoom - please contact emilie_campanelli@fas.harvard.edu for more information

Headshot of Yves AtchadeTitle:

A computational framework for large scale Bayesian inference

Abstract:

The talk deals with statistical models where the number of parameters in the model and/or the number of data points are very large. I will describe a simple spike-and-slab framework modeling strategy, and a novel approximate MCMC sampler to deal with the resulting posterior distribution. The approximate sampling algorithm is remarkably accurate as I will show with several examples. I will also describe some new results on the mixing times of the algorithm in the high-dimensional regime.