Presented by: 
John Ormerod (University of Sydney)
Date: 
Mon 19 Aug, 2:00 pm - 2:45 pm
Venue: 
Otto Hirschfeld Building 81-214

Joint work with Chong You, Samuel Muller and Matthew Stephens

We develop three approaches to the selection of linear models using mean field
variational Bayes (VB) approximations. These include two variational
information criterion and two different VB approximations of linear
regression models using spike and slab priors. Some theory, advantages and
disadvantages of each approach will be discussed. Empirically these methods are
shown to compare favourably with several popular model selection methods both in
terms of computational efficiency and model selection performance. Some preliminary results
where n (the number of samples) is smaller than p (the number of predictors) will also be presented.