This talk has captions. You can remove these by pressing CC on the video toolbar.
Name: Pierre Nyquist
Talk Title: A large deviations analysis of piecewise deterministic Markov processes for MCMC
Abstract: Rare-event sampling is a hindrance to efficiently sample from Gibbs measures, especially in the settings of high dimension or low temperature, where the prevalence of
(deep) local minima causes standard algorithms to converge slowly, at times rendering
them useless. The standard tool for sampling from such measures is Markov chain Monte Carlo (MCMC) methods and constructing samplers that can overcome this difficulty related to rare-event sampling is an active area of research. In this talk I will discuss using tools related to rare events of the underlying empirical measure, specifically large deviation results, to analyse MCMC methods. After a general discussion of this approach, I will focus on MCMC algorithms based on piecewise deterministic Markov processes, a class of methods currently receiving a lot of attention due to their potential benefits, e.g. in the setting of large data sets.
The talk is based on joint work with Joris Bierkens and Mikola Schlottke.
This talk is an invited talk at EVA 2021. View the programme here.