Reduced variance Monte Carlo for Bayesian models with intractable likelihoods
Nial Friel (University College Dublin)
Friday 13th November, 2015 15:00-16:00 Maths 204
Abstract
Many statistical models are intractable, in the sense that the likelihood function cannot easily be evaluated, even up to proportionality. Bayesian estimation in this setting remains challenging. In this paper we construct novel control variates for intractable likelihoods that can reduce the Monte Carlo variance of Bayesian estimators, in some cases dramatically. We prove that our control variates are well-defined, provide a positive variance reduction and derive optimal tuning parameters that are targeted at maximising this variance reduction. Moreover, the methodology is highly parallelisable and offers an alternative route to exploit multi-core processing architectures for Bayesian computation. We illustrate the performance of our methodology on a variety of problems
Add to your calendar
Download event information as iCalendar file (only this event)