Seminars

High-dimensional variable selection

115
reads

Yu-Tin Lin

2012-03-23
13:00:00 - 14:40:00

103 , Mathematics Research Center Building (ori. New Math. Bldg.)

In this talk we will present the work “High-dimensional variable selection” by Wasserman and Roeder (2009). A multi- stage regression method, which is called "screen and clean", is proposed to do variable selection for high-dimensional models. At the first stage a set of candidate models are fitted. At the second stage a most feasible model is selected among the candidate models by cross-validation. These two stages are called "screening". At the third stage a traditional hypothesis testing is used to eliminate insignificant variables. This last stage is called "cleaning". In particular, this screen-and-clean method controls the type-I error rate and establishes variable selection consistency under certain conditions. An application of screen-and-clean to genomic data analysis can be found in Wu et al. (2010).