|
Activity Number:
|
194
|
|
Type:
|
Contributed
|
|
Date/Time:
|
Monday, August 4, 2008 : 2:00 PM to 3:50 PM
|
|
Sponsor:
|
IMS
|
| Abstract - #300891 |
|
Title:
|
G-PLUS Algorithm for Concave-Penalized Negative Likelihood Selection
|
|
Author(s):
|
Wenhua Jiang*+ and Cun-Hui Zhang
|
|
Companies:
|
Rutgers, The State University of New Jersey and Rutgers, The State University of New Jersey
|
|
Address:
|
Department of Statistics, Piscataway, NJ, 08854,
|
|
Keywords:
|
LASSO ; SCAD ; penalized likelihood ; solution path ; variable selection
|
|
Abstract:
|
Penalized methods are widely used for variable selection in high-dimensional data analysis. Concave-penalized negative likelihood approach like the SCAD has caught much attention since it enjoys promising theoretical properties. In this talk we discuss generalized PLUS (G-PLUS) algorithm to compute the solution path of concave-penalized negative likelihood. The G-PLUS is able to discover the possibly multiple local minimizers of an individual penalty level by continuously tracing the minimizers of other different penalty levels. We use numerous end-to-end short linear segments to approximate the nonlinear paths of generalized linear models. Our simulation results show that both the SCAD and the minimax concave penalty (MCP) methods remarkably outperform LASSO in linear and logistic regression.
|
