A1221
Title: Preconditioning in Markov chain Monte Carlo
Authors: Samuel Livingstone - University College London (United Kingdom) [presenting]
Abstract: The purpose is to discuss the quantification of the effectiveness of linear preconditioning in MCMC. Preconditioning is an attempt to modify a target distribution so that it is more amenable to sampling. Linear preconditioning is the most common choice and refers to the act of pre-multiplying the state vector by a constant matrix. Recent results on mixing times in MCMC are leveraged to show scenarios in which commonly used preconditioners will and will not improve sampling. Further discussion on designing subquadratic linear preconditioners that can perform well in the presence of high correlation may be made.
