Skip to main navigation Skip to main content
The University of Southampton
Southampton Statistical Sciences Research Institute

Bayesian variable selection in generalised linear models using a combination of Stochastic Optimisation Methods Seminar

Date:
28 September 2012
Venue:
Building 39 Room 3013

For more information regarding this seminar, please email Mrs Jane Revell at j.revell@southampton.ac.uk .

Event details

Methodology seminar

Abstract
The talk consists of two parts. In Part 1 a description of how the Bayesian community deals with the variable selection problem will be presented. Several popular approaches to Bayesian variable selection, including computational methods for posterior evaluation and exploration, will be briefly reviewed. Additionally, different approaches on specifying the prior distribution (a) on the parameter space of each candidate model and (b) on the model space, will be shown. In Part 2 the usage of a stochastic optimisation algorithm as a model search tool will be proposed for the Bayesian variable selection problem in generalised linear models. Combining aspects of three well known stochastic optimisation algorithms, namely, simulated annealing, genetic algorithm and tabu search, a powerful model search algorithm will be produced. After choosing suitable priors, the posterior model probability will be used as a criterion function for the algorithm; in cases when it is not analytically tractable Laplace approximation will be used. The proposed algorithm will be illustrated on normal linear and logistic regression models, for simulated and real-life examples, and it will be shown that, with a very low computational cost, it achieves improved performance when compared with popular MCMC algorithms, such as the MCMC model composition, as well as with vanilla versions of simulated annealing, genetic algorithm and tabu search.

Speaker information

Professor Dimitris Fouskakis , School of Applied Mathematical and Physical Sciences, Athens. Assistant Professor

Privacy Settings