Colloquium-Aaron Smith (Tutte Institute of Mathematics and Computation)-Convergence of Approximate Adaptive Samplers

Mathematics - Colloquium

Thursday, January 23, 2014
11:00 AM-12:00 PM

Stratton Hall
203

ABSTRACT: Markov chain Monte Carlo (MCMC) is a standard tool for sampling from complicated probability distributions. In general, the implementation of MCMC algorithms requires the choice of a large number of tuning parameters, and this choice can have an enormous impact on the efficiency of the algorithm. In practice, this problem is often resolved by using "adaptive" MCMC algorithms, which automatically adjust these tuning parameters as they run. I will introduce some MCMC algorithms, some well-known adaptive variants, and present recent work providing the first proofs that the adaptive variants are more efficient than the original algorithms. I will also briefly discuss the related problem of running MCMC for infereince problems involving massive amounts of data.

Suggested Audiences: College, Adult

E-mail: ma-chair@wpi.edu

Last Modified: December 17, 2013 at 3:18 PM