CS Forum: Richard Everitt, University of Reading "ABC and Synthetic Likelihood for Expensive Simulators"
ABC and Synthetic Likelihood for Expensive Simulators
University of Reading
Approximate Bayesian computation (ABC) is now an established technique for statistical inference in the form of a simulator, and approximates the likelihood at a parameter θ by simulating auxiliary data sets x and evaluating the distance of x from the true data y. Synthetic likelihood is a related approach that uses simulated auxiliary data sets to construct a Gaussian approximation to the likelihood. The talk will begin by reviewing and comparing these approaches. Neither approach is computationally feasible in cases where using the simulator for each θ is very expensive, without some algorithmic refinements. This talk investigates using the bootstrap to cheaply estimate the synthetic likelihood from a single sample from the likelihood. We also examine a synthetic likelihood approximation that is constructed, using the bag of little bootstraps, from subsampled data sets. Applications to stochastic differential equation models and doubly intractable distributions will be presented. The work in this talk is described further in the paper “Bootstrapped synthetic likelihood”, arxiv.org/abs/1711.05825 (and will also touch on material from the papers at arxiv.org/abs/1708.02230, https://bit.ly/2Hgutbf and https://bit.ly/2TRc5LM).
Richard Everitt is an Associate Professor at the University of Reading, in the Department of Mathematics and Statistics. He began his career in industry, and during this time studied for a PhD at the University of Bristol with Peter Green and Christophe Andrieu. He then joined Bristol as a Brunel Fellow, then Oxford as a postdoc working in genomics. He joined Reading in 2012. His work is in computational statistics and its applications, including to problems in genetics, ecology, climate and neuroscience.
CS forum is a seminar series arranged at the CS department. The talks are intended for presentations of postdoctoral level researchers and professors, both for visiting and CS-department-based researchers.