Oracle Inequalities for Structured Variable Selection by Multi-Regularization
主 题: Oracle Inequalities for Structured Variable Selection by Multi-Regularization
报告人: Dr. Yiyuan She, Associated Professor（Department of Statistics, Florida State University）
时 间: 2014-05-21 16:00-17:00
地 点: 理科一号楼1303会议室（统计中心活动）
Recently many multi-regularized estimators have been proposed and applied to capture various types of structural parsimony in high-dimensional applications. Yet with multiple sparsity or low rank promoting penalties enforced on the same object, the statistical analysis is difficult, and there is a lack of finite-sample studies in the literature. In addition, with multiple regularization parameters involved, the parameter tuning is quite messy and ad-hoc. This talk presents some nonasymptotic results based on two novel recipes. L1 type penalties as well as nonconvex penalties in sparse and/or low rank problems can all be handled to give sharp rates. Some examples are demonstrated to show the efficacy and efficiency of the proposed methodologies.
