"Markov-Chain Monte Carlo" (MCMC) is a technique for generating random samples from a specified probability distribution --- often one of a combinatorial or statistical-mechanical nature --- by simulating a Markov chain whose state space includes the structures of interest. MCMC methods are widely applied in diverse areas such as computational biology, astronomy, finance, statistics, computer science and statistical physics. MCMC methods are only applicable when the Markov chain converges rapidly to the target distribution. The mathematical analysis of MCMC methods, focusing on the rate of convergence to equilibrium, has become a highly developed branch of probability theory and theoretical computer science. Lately, ideas from statistical physics and elsewhere have led to a number of variants and alternatives to the MCMC methodology. The study of these alternatives requires the development of new mathematical techniques.
This workshop aims to bring together researchers from all the above-mentioned areas with the goal of deepening cooperation and promoting the cross-fertilization of ideas.
Topics will include (but are not limited to):
- The context: models, problems and ideas from physics and elsewhere.
- Probabilistic and analytic techniques for MCMC (sophisticated couplings, Martingale methods, log-Sobolev, cutoff).
- Developments of MCMC, and competing techniques.
- Computational perspectives (including fundamental complexity-theoretic barriers).
Keynote Speakers will include
Eric Vigoda (Georgia Tech), Fabio Martinelli (Rome) and Prasad Tetali (Georgia Tech).