COLLOQUIUM

DEPARTMENT OF MATHEMATICS AND STATISTICS

OAKLAND UNIVERSITY

ROCHESTER, MICHIGAN  48309

Xiaoli Gao

Department of Mathematics and Statistics

Oakland University

Rochester, Michigan 48309

The LAD Adaptive Lasso in a High-dimensional Sparse Model

Abstract

High-dimensional data have been involved in many important technologies and applications.  The extraction of relatively few but true information from many complicated features is very important.  The least absolute shrinkage and selection operator (Lasso)  becomes a very popular approach in a sparse high-dimensional setting.  However, the Lasso tends to over-select models in general since it penalizes all the covariates equally.  The adaptive Lasso outperforms the Lasso in terms of its oracle properties.  In this talk, I will discuss some robust and oracle properties of penalized least absolute deviations  regression via adaptive Lasso (LAD-AL) in a high-dimensional sparse model.  The finite sample performance of LAD-AL and its extensions will be demonstrated by simulation studies and some statistical applications in genetics.

Thursday, November 19, 2009

3:00 – 4:00 P.M.

372 Science and Engineering Building

(Refreshments at 2:30-3:00 PM in the kitchen area adjacent to 368 SEB)