Скачать или смотреть Grigorios A Pavliotis: Accelerating convergence and reducing variance for Langevin samplers

Grigorios A. Pavliotis: Accelerating convergence and reducing variance for Langevin samplers

Markov Chain Monte Carlo (MCMC) is a standard methodology for sampling from probability distributions (known up to the normalization constant) in high dimensions.

There are (infinitely) many different Markov chains/diffusion processes that can be used to sample from a given distribution. To reduce the computational complexity, it is necessary to consider Markov chains that converge as quickly as possible to the target distribution and that have a small asymptotic variance. In this talk I will present some recent results on accelerating convergence to equilibrium and on reducing the asymptotic variance for a class of Langevin-based MCMC algorithms.

The lecture was held within the framework of the Hausdorff Trimester Program Optimal Transportation and its Applications. (28.01.2015)