Anthony Lee
Thursday 16th April 2015
Time: 3.30pm
Basement Seminar Room
Alexandra House, 17 Queen Square, London, WC1N 3AR
Scalable Monte Carlo for Complex Models
As we acquire larger data sets, we are increasingly drawn to complex probabilistic models to explain the mechanisms by which they were produced. While theoretically very exciting, fitting such models with data presents fundamental computational challenges. One challenge is that realistic models are often naturally intractable, complicating the design of traditional Monte Carlo inference algorithms. Another challenge is to design algorithms that scale on parallel and distributed architectures. I will discuss contributions from two projects addressing each of these challenges in particular settings: an efficient Markov chain Monte Carlo algorithm for approximate Bayesian computation and a generalization of sequential Monte Carlo that is amenable to parallel and distributed implementation.