Title
Estimating ocean circulation: a likelihood-free MCMC approach via a Bernoulli factory
Speaker
Radu Herbei, The Ohio State University
Abstract
We describe a novel Markov chain Monte Carlo approach that does not require a likelihood evaluation. Rather, we use unbiased estimates of the likelihood and a Bernoulli factory to decide whether proposed states are accepted or not. We illustrate this approach using a oceanographic data inversion example. The variates required to estimate the likelihood function are obtained via a Feynman-Kac representation. This lifts the restriction of selecting a regular grid for the physical model and eliminates the need for data pre-processing. We implement our approach using the parallel GPU computing environment.
Radu Herbei is originally from Romania. He has a BS degree from West University, Timisoara, double-majoring in mathematics and finance. He got his PhD from the Statistics Department at Florida State University and joined Ohio State in the fall of 2006. His research interests lie in the general area of applied probability. In the recent past, he has been studying theoretical and computational aspects arising in MCMC procedures, such as perfect sampling algorithms for general distributions and rates of convergence for Markov chains on very large finite state spaces.
