A0819
Title: Group sparse variational Bayes approach for high-dimensional data
Authors: Sarah Filippi - Imperial College London (United Kingdom) [presenting]
Michael Komodromos - Imperial College London (United Kingdom)
Kolyan Ray - Imperial College London (United Kingdom)
Marina Evangelou - Imperial College London (United Kingdom)
Abstract: Few Bayesian methods for analyzing high-dimensional sparse data provide scalable variable selection, effect estimation and uncertainty quantification. Most methods either sacrifice uncertainty quantification by computing maximum a posteriori estimates or quantify the uncertainty at high (unscalable) computational expense. The focus is on a method for the selection of groups of variables under a generalized linear model. For this setting, an interpretable and scalable model is developed for prediction and group selection. The method, based on a variational approximation, overcomes the high computational cost of MCMC whilst retaining the useful features, providing excellent point estimates and offering a natural mechanism for group selection via posterior inclusion probabilities. Posterior concentration rates are derived for the group sparse variational Bayes approach. They compare the methods against other state-of-the-art Bayesian variable selection methods on simulated data and demonstrate their application for variable and group selection on real biomedical data.
