【学术报告】Variable selection via a concave amalgamated penalty

发布者:系统管理员发布时间:2015-11-02浏览次数:32

学术报告


报告题目:Variable selection via a concave amalgamated penalty

报 告 人:王学钦博士,中山大学数学与计算科学学院和中山医学院双聘教授、国家优秀青年基金获得者

时 间:2015114(星期三)上午9:30

地 点:数学楼第七教室

摘 要: A new regularization method is proposed based on a concave amalgamated penalty (CAP), which simultaneously selects variables and estimates coefficients in ultra-high dimensional cases, especially when the number of the candidate variables p is much greater than the number of observations n. The CAP estimator retains the grouping effect property where strongly correlated predictors tend to be in or out of the model together. By proper control of the tuning parameters, we prove that the CAP estimator achieves global optimality when both n and p converge to infinity, i.e., the CAP estimator is asymptotically equivalent to the "oracle" estimator, and thus possesses the oracle property. The CAP is efficiently implemented using a modified coordinate gradient descent algorithm. In addition, we present several simulation studies and real data analysis to illustrate and provide insight into the proposed method.

‍‍欢迎广大师生参加!

统计研究院

2015112