主 题：Robust Conditional Sure Independence Screening via Blum-Kiefer-Rosenblatt Correlation learning
内容简介：Marginal screening methods have been widely used in the high dimensional data analysis. Despite they are easy to implement, they still suffer from the failure in detecting the important predictors with weak marginal signals. In this paper we develop a model-free conditional screening procedure based on conditional Blum-Kiefer-Rosenblatt correlation (CBKR for short), a metric to measure the conditional contributions of predictors to the response. Our proposed procedure is robust to the presence of extreme values and outliers in the observations, indicating it can accommodate the heterogeneity in the high dimensional data. We also show that, under mild conditions, the proposed procedure has the desirable sure screening property,which guarantee that all important predictors can be retained after screening with probability approaching one. Moreover, we provide a data-driven procedure to determine the number of features to be retained after screening. The usefulness of this conditional screening procedure is illustrated by the simulation studies and an application to the gene expression microarray dataset of rat eye.
报告人：朱利平 教授 博导 国家优青
时 间：2017-04-28 14:00
举办单位：理学院 统计科学与大数据研究院 科研部