报 告 人： 安百国
报告地点： 数学与统计学院105室
报告时间： 2016年03月18日星期五16:00-17:00
主讲人简介：
安百国，首都经济贸易大学统计学院讲师，2012年博士毕业于东北师范大学数学与统计学院，师从郭建华教授；2013-2015年先后在北卡罗来纳大学统 计与运筹系、生物统计系进行博士后访问，2012-至今在首都经济贸易大学统计学院做博士后工作，合作导师是纪宏教授。主要的研究兴趣包括回归压缩与选 择、机器学习、超高维数据分析、文本挖掘和神经影像学分析等。
报告简介： The aim of this paper is to develop a class of linear mixed effects models with functional covariates (LMMFC). This development is motivated by the use of neuroimaging marker(s) to predict longitudinal clinical outcomes. A total variation penalty is introduced to characterize slope image(s) associated with the functional covariates in LMMFC. An Alternating Direction Method of Multipliers (ADMM) algorithm is proposed to simultaneously estimate the slope image and other unknown parameters, such as variance components in LMMFC. Theoretically, we establish an upper bound for the estimation error and prove the consistency of the variance component estimates. We also develop a bootstrap test procedure to carry out a global significance test for the slope image. Our simulations demonstrate a superior performance of LMMFC over many existing approaches. We use LMMFC to predict clinical scores by using a hippocampus covariate obtained from the Alzheimers Disease Neuroimaging Initiative (ADNI) study.
