Variational approximations for mixed models

Variational approximations are a general class of methods designed to produce a closed form approximation to an otherwise intractable marginal likelihood. The project will review the basic concepts and current statistical literature on variational approximations for mixed and more general multilevel models, perform empirical evaluations of variational approximations versus other methods of estimation in different cases, and potentially consider extensions to inference.

This project will be co-supervised with Dr Francis Hui.