学术报告

题目：An optimality of the L_0 Penalty and   dual achievability of model selection

报告人：Yuhong Yang

              (University of Minnesota)

Abstract :  Various penalization methods have been proposed for high-dimensional learning to achieve a sparse subset selection. A fundamental question is: Which penalty form is the best and in what sense? In this talk, we present a formal result on optimality of the L_0 penalty for variable selection in terms of the under-fitting and over-fitting probabilities. It is used to obtain a precise understanding on dual achievability of model selection, i.e., if and when we can achieve both model selection consistency and minimax rate estimation of the regression function in strong and weak senses. The work is joint with Zhan Wang.

时间：4月26日（周五）下午14:30-15：30

地点：565net必赢客户端本部教二楼 513 教室

联系人：邹国华

欢迎全体师生积极参加！