Talk Title

Beyond Asymptotically Efficient Markov Chains

Abstract

Two commonly acknowledged weaknesses of MCMC methods are that they can be computationally very intensive (compared to say variational methods or ABC) and that their convergence bounds are overly conservative. Historically, the research has focused on finding new Markov chains that scale better with the difficulty of a given statistical problem and have better convergence rate. Recently, there has been a surge of interest for Markov chains whose sub-optimal asymptotic convergence rate is balanced out by other appealing features which address (in some ways) the aforementioned traditional weaknesses. The talk will present such MCMC algorithms and discuss some of their properties.  

Bio

Florian Maire is an Assistant Professor at the University of Montreal, Canada, since 2018. Prior to that, he got is PhD from Telecom SudParis and University Paris 6 in 2014 and was a research fellow at University College Dublin, Ireland. His research area is on the design and analysis of methods used in Statistical Learning, such as Markov chain Monte Carlo algorithms in Bayesian Statistics (MCMC). His current work includes exact approximations, scalability and complexity of MCMC methods with applications to high-dimensional problems.