Bayesian Sparse Group Selection

Abstract

Bayesian Sparse Group Selection Ray-Bing Chen Department of Statistics, National Cheng Kung University, Taiwan ?: Consider the linear models with grouped variables. Here we are not only interested in selecting the proper groups but also want to simultaneously identify the active variables in selected groups. Two Bayesian selection approaches are proposed. One approach is directly generalized from the componentwise Gibbs sampler for Bayesian variable selection. Another one is a Metropolis type algorithm. Several simulations are used to demonstrate the performances of the proposed approaches. The simulation results show that both Bayesian selection approaches are compatible with the sparse group lasso proposed by Simon, Friedman, Hastie and Tibshrani (2012). This is a joint work with Ying Nian Wu (UCLA) and Chi-Hsiang Chu (NUK).