Events login

Main Events Calendar

Warning Icon This event is in the past.
January 19, 2018 | 2:45 p.m. - 4:00 p.m.
Category: Lecture
Location: Education, College of #169 | Map
5425 Gullen Mall
Detroit, MI 48202
Cost: Free

Speaker: Songshan Yang

Title: Sure Joint Ranking and Screening in Ultrahigh Dimensional Linear Regression Models

Abstract: Feature screening is fundamental for analysing the ultrahigh dimensional data, where the number of predictors is much larger than the sample size. This paper develops a new feature screening procedure, sure joint ranking and screening, for linear regression models with ultrahigh dimensional covariates. A joint utility measure is proposed for predictor ranking, which is motivated by the fact that important predictors tend to enter the model before unimportant ones on the solution path of penalized regressions. The new feature screening procedure considers the joint effects between features and possesses consistency in ranking, which gives itself an advantage over other existing methods. The cutoff for separating the important and unimportant predictors is obtained by adding permuted predictors as pseudo-variables. Our simulation indicates that the new joint screening procedure outperforms other existing feature screening methods for the purpose of sure screening and model selection consistency. We also illustrate the proposed feature screening procedure by an empirical analysis of a Chinese supermarket data set.

For more information about this event, please contact Department of Mathematics at 3135772479 or math@wayne.edu.