Now showing 1 - 2 of 2
  • Publication
    Exploiting Multi-Core Architectures for Reduced-Variance Estimation with Intractable Likelihoods
    (International Society for Bayesian Analysis (ISBA), 2015) ; ;
    Many popular statistical models for complex phenomena areintractable, in the sense that the likelihood function cannot easily be evaluated.Bayesian estimation in this setting remains challenging, with a lack of computa-tional methodology to fully exploit modern processing capabilities. In this paperwe introduce novel control variates for intractable likelihoods that can dramati-cally reduce the Monte Carlo variance of Bayesian estimators. We prove that ourcontrol variates are well-defined and provide a positive variance reduction. Fur-thermore, we show how to optimise these control variates for variance reduction.The methodology is highly parallel and offers a route to exploit multi-core pro-cessing architectures that complements recent research in this direction. Indeed,our work shows that it may not be necessary to parallelise the sampling processitself in order to harness the potential of massively multi-core architectures. Simu-lation results presented on the Ising model, exponential random graph models andnon-linear stochastic differential equation models support our theoretical findings.
      349ScopusĀ© Citations 13
  • Publication
    Investigation of the widely applicable Bayesian information criterion
    The widely applicable Bayesian information criterion (WBIC) is a simple and fast approximation to the model evidence that has received little practical consideration. WBIC uses the fact that the log evidence can be written as an expectation, with respect to a powered posterior proportional to the likelihood raised to a power t(0,1)t(0,1) , of the log deviance. Finding this temperature value tt is generally an intractable problem. We find that for a particular tractable statistical model that the mean squared error of an optimally-tuned version of WBIC with correct temperature tt is lower than an optimally-tuned version of thermodynamic integration (power posteriors). However in practice WBIC uses the a canonical choice of t=1/log(n)t=1/log(n) . Here we investigate the performance of WBIC in practice, for a range of statistical models, both regular models and singular models such as latent variable models or those with a hierarchical structure for which BIC cannot provide an adequate solution. Our findings are that, generally WBIC performs adequately when one uses informative priors, but it can systematically overestimate the evidence, particularly for small sample sizes.
      355ScopusĀ© Citations 20