A0674
Title: A new group variable screening approach for ultrahigh-dimensional data
Authors: Daoji Li - California State University Fullerton (United States) [presenting]
Abstract: In many applications with ultrahigh-dimensional data, variables are naturally grouped. There is a rich literature on variable screening, but most of the existing work does not consider the grouping structure. We propose a new group screening approach that allows predefined groups of predictors to be identified to address this issue jointly. The proposed method enjoys the sure screening property and does not require any assumption on the distribution of variables. Various numerical studies confirm the superior empirical performance of the proposed methods.
