Adaptive MCMC and model selection
Document Type
Presentation Abstract
Presentation Date
3-9-2009
Abstract
Several adaptive Markov chain Monte Carlo (MCMC) methods for doing Bayesian statistical inference have been developed recently. These adaptive methods aim to increase the efficiency of the Metropolis-Hastings sampling algorithm and to allow automatic tuning of the algorithm's proposal distribution. This is useful especially in estimation problems in high dimensional and computationally intensive models. Adaptive methods are also good building blocks for implementation of general MCMC software as they need very little user intervention compared to the non-adaptive versions. Theoretical considerations are needed to ensure that the adaptive methods preserve the correct stationary distribution. Application examples are presented in the fields of environmental modeling and geophysical remote sensing. An extension to Adaptive Metropolis called adaptive automatic reversible jump MCMC (AARJ) allows for Bayesian model selection and averaging, and to incorporate model uncertainty in the statistical analysis of nonlinear models.
Recommended Citation
Laine, Marko, "Adaptive MCMC and model selection" (2009). Colloquia of the Department of Mathematical Sciences. 309.
https://scholarworks.umt.edu/mathcolloquia/309
Additional Details
Monday, 9 March 2009
3:10 p.m. in Math 103
4:00 p.m. Refreshments in Math Lounge 109