Bayesian Numerical Methods
Professor Christian Robert
CREST and University of Dauphine, Paris
Thursday 19th and Friday 20th July, 9.30am to 1.00pm
Synopsis
The aim is to cover the recent emergence of dedicated numerical methods for processing the ever-increasing complexity in statistical models. Time-series models, latent variable models, and untractable likelihood models are three specific categories of (potentially) problematic models that have led to the emergence of specialised responses: sequential Monte Carlo (SMC) methods and particle filters, Markov chain Monte Carlo (MCMC) methods, and approximate Bayesian computation (ABC) methods. I plan to cover those methods at the methodological level, illustrating their implementation in the above categories, rather than embarking upon a complete theoretical justification.