题目: Response best-subset selector for multivariate regression with large-scale response variables

报告人：刘旭

时间：10.26日（星期三），13:00-14:00

地点：线下：综合楼644

          线上：腾讯会议894 711 441

摘要: This article investigates the statistical problems of response variable selection with exponentially large-scale response variables for multivariate linear regression with two settings of fixed and diverging numbers of predictor variables. A response best-subset selection model is proposed by introducing a 0-1 selection indictor to each response variable, and the response best-subset selector is developed, which is an efficient procedure for performing response variable selection and regression coefficient estimation simultaneously by introducing a separation parameter and a novel penalty function. The proposed response best-subset selectors for two settings both have consistency under mild conditions. For the fixed number of predictor variables, consistency and asymptotic normality are presented for the corresponding regression coefficient estimators. The Bonferroni test procedure with F-type test statistics turns out to be a special case of our response best-subset selector. In finite-sample simulation studies our response best-subset selector has stronger competitive advantages in keeping balance between higher accurate rates of important and unimportant response variables or larger Matthews correlation coefficient over its main competitors. A real data analysis demonstrates the effectiveness of the response best-subset selector for identifying dosage-sensitive genes.

报告人简介：

刘旭博士是上海财经大学统计与管理学院副教授，博士生导师。2011年博士毕业于云南大学。2011-2013年在美国西北大学做博士后研究，2013-2016年在密歇根州立大学做博士后研究。研究兴趣为机器学习，Tensor统计建模，高维数据和基因数据分析，以及非参半参数统计建模。在国际权威统计期刊包括JASA，Biometrika，Biometrics，Statistica Sinica等发表近20篇论文。主持两项国家自然科学基金面上项目。