Title: Variational Bayesian approaches for Structure Selection
Authors: Lai Wei-Ting - National Central University (Taiwan) [presenting]
Abstract: The problems of group structure selection in high-dimensional regression models are considered. The variables or regressors are partitioned into different group and only a few groups are active or important. Therefore, it is an interesting issue finding important groups. The Bayesian approach based on the spike and slab priors for the regression coefficients is considered, and each candidate group is associated with a binary variable indicating whether the group is active or not. Instead of Markov Chain Monte Carlo (MCMC) methods, a fast and scalable vibrational Bayesian approach is introduced for the posterior inference. Furthermore, we extend the proposed method for the multi-task learning and the structure selection problem in vector autoregression (VAR) model. Simulation studies and real examples are used to demonstrate the performances of the proposed Bayesian approaches.