|
Activity Number:
|
1
|
|
Type:
|
Invited
|
|
Date/Time:
|
Sunday, July 29, 2007 : 2:00 PM to 3:50 PM
|
|
Sponsor:
|
Section on Bayesian Statistical Science
|
| Abstract - #307899 |
|
Title:
|
Covariance Selection and Bayes Classification via Modal Shrinkage Estimators
|
|
Author(s):
|
Jingqin Luo*+ and Merlise A. Clyde and Edwin Iversen
|
|
Companies:
|
Washington University in St. Louis and Duke University and Duke University
|
|
Address:
|
Box 8067, Division of Biostat, School of Medicine, St. Louis, MO, 63110,
|
|
Keywords:
|
Bayesian analysis ; covariance selection ; Baye classification ; shrinkage regression ; scale mixtures of normals
|
|
Abstract:
|
Due to the positive definiteness constraint and the rapidly growing number of parameters with dimensions, covariance estimation in a multivariate normal population has been a classic but challenging statistical problem. Many approaches shrink a covariance/precision matrix toward some special parsimonious structures, which may suffer from misspecification error. By describing the covariance selection problem as a system of linear recursive equations, we work in the Cholesky decomposition framework of a precision matrix. Through application of Bayesian shrinkage regressions, we obtain robust estimators for a precision matrix of a flexible sparse pattern. A further application of Bayesian shrinkage regressions to Bayes classifier results in classifications comparable to some state-of-the-art methods.
|
