Monte Carlo Approximation of Bayes Factors via Mixing with Surrogate Distributions
share
By mixing the posterior distribution with a surrogate distribution, of which the normalizing constant is tractable, we describe a new method to estimate the normalizing constant using the Wang-Landau algorithm. We then introduce an accelerated version of the proposed method using the momentum technique. In addition, several extensions are discussed, including (1) a parallel variant, which inserts a sequence of intermediate distributions between the posterior distribution and the surrogate distribution, to further improve the efficiency of the proposed method; (2) the use of the surrogate distribution to help detect potential multimodality of the posterior distribution, upon which a better sampler can be designed utilizing mode jumping algorithms; (3) a new jumping mechanism for general reversible jump Markov chain Monte Carlo algorithms that combines the Multiple-try Metropolis and the directional sampling algorithm, which can be used to estimate the normalizing constant when a surrogate distribution is difficult to come by. We illustrate the proposed methods on several statistical models, including the Log-Gaussian Cox process, the Bayesian Lasso, the logistic regression, the Gaussian mixture model, and the g-prior Bayesian variable selection.
READ FULL TEXT
