学术报告
报告题目：On Marginal Sliced Inverse Regression for Ultrahigh Dimensional Feature Selection
报告人：於州 副教授
华东师范大学
时间：2016年3月31日8:30-9:30
地点：统计研究院426教室
摘要: Model-free variable selection has been implemented under the sufficient dimension reduction framework since the seminal paper ofCook (2004). In this paper, we extend the marginal coordinate testfor sliced inverse regression (SIR) in Cook (2004) and propose a novelmarginal SIR utility for the purpose of ultrahigh dimensional featureselection. Two distinct procedures, Dantzig selector and sparse precision matrix estimation, are incorporated to get two versions of samplelevel marginal SIR utilities. Both procedures lead to model-free variable selection consistency with predictor dimensionality p divergingat an exponential rate of the sample size n. As a special case ofmarginal SIR, we ignore the correlation among the predictors andpropose marginal independence SIR. Marginal independence SIR isclosely related to many existing independence screening proceduresin the literature, and achieves model-free screening consistency in theultrahigh dimensional setting. The _nite sample performances of theproposed procedures are studied through synthetic examples and anapplication to the small round blue cell tumors data.
欢迎广大师生参加！
统计研究院
2016年3月18日