B1070
Title: Scalable Bayesian estimation of sparse Gaussian graphical models
Authors: Deborah Sulem - Barcelona School of Economics (Spain) [presenting]
David Rossell - Universitat Pompeu Fabra (Spain)
Jack Jewson - Universitat Pompeu Fabra and Barcelona Graduate School of Economics (Spain)
Abstract: Gaussian graphical models are widely used to analyse the conditional dependence structure between variables such as gene expression data. Under the Gaussian model, inferring this dependence structure corresponds to estimating a precision matrix, often high-dimensional when the number of variables considered in the analysis, p, is large. When p is big, potentially larger than the sample size n, it is often reasonable to assume that the precision matrix is sparse and the graphical model is parsimonious, since the zero entries imply the conditional independence property. Then only the most significant partial dependencies are sought to be estimated. A fast Bayesian method is proposed to estimate a precision matrix and infer a graphical model in high-dimensional settings. In this context, estimating the posterior distribution via Monte-Carlo Markov Chains remains challenging, because of the size of the model space which grows exponentially with the number of variables. The approach consists of parallelising the computations of approximated posterior conditional distributions, which enables a fast exploration of the local partial correlation structure.
