报告题目:
ROBUST SUBGROUP IDENTIFICATION
稳健子群识别
报告时间:2019年4月16日(星期二)上午9:30-10:30
报告地点:澳门十大娱乐平台正规二楼会议室
报告人:朱仲义,复旦大学统计系教授,博士生导师。曾任中国概率统计协会第八,九届副理事长;国际著名杂志“Statistica Sinica” 副主编;“应用概率统计”,“数理统计与管理”杂志编委;中国统计教材编审委员会委员;现为Elected Member of ISI(国际数理统计协会);“中国科学.数学”杂志编委;研究方向包括:保险精算,纵向数据模型,分位数回归模型等。主持完成国家自然科学基金4项,国家社会科学基金1项,作为子项目负责人完成国家自然科学基金重点项目1项。目前主持国家自然科学基金重大项目子项目1项,重点项目子项目1项,面上项目1项。近年来发表论文100多篇,其中包括国际统计类四大顶级杂志SCI论文60多篇。第一完成人获中华人民共和国教育部自然科学二等奖1项。
报告摘要:
We consider a more robust subgroup analysis based on median regression and model the heterogeneity by subject-specific intercepts. This model have two appealing featurs: (1) By considering median regression, we allow the errors to have heavy tails or correlation with covariates. (2) The characteristic of quantile loss function enables us to develop augmented dataset to implement Local Linear Approximation Algorithm. In this paper, we investigate nonconvex pairwise penalized median regression for simultaneous subgroup analysis and parameter estimation. To derive the oracle property, we employ a recently developed convex-differencing method to tackle the challenges of nonsmooth loss function and nonconvex penalty functions. Simulation results and an application to the Cleveland heart disease dataset demonstrate the effectiveness of the proposed method.