Department Colloquim: Steffen Lauritzen
Title: Gaussian Graphical Models with Symmetry.
Speaker: Steffen Lauritzen, Oxford University.
Abstract: Gaussian Graphical Models are statistical models describing dependence structures among variables in terms of conditional independence encoded in a Markov property. They are generally determined by an undirected graph where missing edges in the graph correspond to zero entries in the inverse covariance matrix.
For reasons of parsimony, and other reasons, it is of interest to study such models which obey additional symmetry restrictions, either determined by specified equality of parameters, or by actions of permutation groups.
I shall describe the basic facts associated with Gaussian graphical models, some of the positive and simple results for special subtypes of models with symmetry, and finally describe a few open questions associated with existence of relevant estimators.