BEGIN:VCALENDAR
VERSION:2.0
PRODID:-//Talks.cam//talks.cam.ac.uk//
X-WR-CALNAME:Talks.cam
BEGIN:VEVENT
SUMMARY: Stepwise Searching for Feature Variables in High-Dimensional Line
 ar Regression -  Qiwei Yao\, LSE
DTSTART:20090130T160000Z
DTEND:20090130T170000Z
UID:TALK16634@talks.cam.ac.uk
CONTACT:8419
DESCRIPTION:We investigate the classical stepwise forward and backward\nse
 arch methods\nfor selecting sparse models in the context of linear regress
 ion with the\nnumber of candidate variables p greater than the number of o
 bservations\nn. Two types of new information criteria BICP and BICC are pr
 oposed to\nserve as the stopping rules in the stepwise searches\, since th
 e\ntraditional information criteria such as BIC and AIC are designed for t
 he\ncases with p<n\, and may fail spectacularly when p is close to or\ngre
 ater than n.  The proposed methods are illustrated in a simulation\nstudy 
 which indicates that the new methods outperform a counterpart LASSO\nselec
 tor.  The consistency of the stepwise search is investigated when\nboth $n
 $ and p tend to infinity. We show that a stepwise forward\naddition follow
 ed by a stepwise backward deletion\, both controlled by a\nversion of BICP
 \, leads to a consistent estimated model under the sparse\nRiesz condition
 .\n
LOCATION:MR12\, CMS\, Wilberforce Road\, Cambridge\, CB3 0WB
END:VEVENT
END:VCALENDAR