A0592
Title: Some insights into the reliability and limitations of adaptive MCMC
Authors: Austin Brown - Texas AM University (United States) [presenting]
Abstract: The reliability and limitations of adaptation strategies used in adaptive Markov chain Monte Carlo (MCMC) are investigated. In particular, general lower bounds are established on the weak convergence rate under general adaptation plans. If the adaptation diminishes sufficiently fast, comparable convergence rate upper bounds are also developed. These results provide some insight into the optimal design of adaptation strategies and help set expectations for the practical performance of adaptive MCMC. Applications to an adaptive unadjusted Langevin algorithm and to adaptively tuning the covariance matrix in random-walk Metropolis-Hastings are explored.
