Towards Not Being Afraid of the Big Bad Dataset

Towards Not Being Afraid of the Big Bad Dataset

Gareth Roberts
Department of Statistics, University of Warwick

Wednesday, Mar.4, 2015
3:00 - 4:00 PM
Sequoia 200

Abstract:

This talk will present the foundations behind a new algorithm for systematic error-free Monte Carlo simulation from intractable target distributions. The main motivation behind the work is to construct a method for exploring posterior distributions for Bayesian analyses of extremely large datasets where computation of the likelihood function at each iteration of an algorithm is prohibitively expensive. The algorithm is a continuous time sequential Monte Carlo procedure which extends many of the ideas used in exact simulation from diffusion sample paths. This is joint work with Paul Fearnhead Murray Pollock and Adam Johansen.

Operations Research Colloquia: http://or.stanford.edu/oras_seminars.html