A0784
Title: Variants of high-dimensional EM algorithm for mixed linear regression
Authors: Ning Wang - Beijing Normal University (China) [presenting]
Abstract: The expectation-maximization (EM) algorithm and its variants are widely used in statistics. In high-dimensional mixture linear regression, the model is assumed to be a finite mixture of linear regression, and the number of predictors is much larger than the sample size. The aim is to consider solving the high-dimensional problem in mixed linear regression. The regression coefficients are assumed to be sparse, and the sparsity pattern is the same for different mixtures. A group lasso penalized approach is presented, a modified group lasso penalized approach, and a subset selection-based approach. The non-asymptotic convergence results for the direct output of the algorithms will be provided. The proposed methods have encouraging performances in numerical studies.
