|
Activity Number:
|
355
|
|
Type:
|
Invited
|
|
Date/Time:
|
Wednesday, August 1, 2007 : 8:30 AM to 10:20 AM
|
|
Sponsor:
|
Section on Statisticians in Defense and National Security
|
| Abstract - #308050 |
|
Title:
|
A Robust Hybrid of Lasso and Ridge Regression
|
|
Author(s):
|
Art Owen*+
|
|
Companies:
|
Stanford University
|
|
Address:
|
Department of Statistics, Stanford, CA, 94025,
|
|
Keywords:
|
Dantzig selector ; lars ; ridge regression ; robustness
|
|
Abstract:
|
A penalty that behaves like lasso for small coefficients and like ridge for large coefficients is developed. This penalty is a reversed Huber function. The penalty is convex. Like the Huber function it requires scaling. The scaling parameter can be incorporated into a criterion that is jointly convex in it and the regression coefficient vector.
|
