|
Activity Number:
|
493
|
|
Type:
|
Invited
|
|
Date/Time:
|
Thursday, August 7, 2008 : 10:30 AM to 12:20 PM
|
|
Sponsor:
|
Section on Statistical Computing
|
| Abstract - #300196 |
|
Title:
|
Adaptive Data Weighting Strategies for Locating the Global Maximum in EM-Type Algorithms
|
|
Author(s):
|
Ravi Varadhan*+
|
|
Companies:
|
Johns Hopkins University
|
|
Address:
|
2024 E. Monument Street, School of Medicine, Baltimore, MD, 21205,
|
|
Keywords:
|
local maxima ; latent class models ; finite mixtures ; EM acceleration ; global maximization ; squared iterative methods
|
|
Abstract:
|
We explore and evaluate some data weighting strategies that adaptively reweigh the data to facilitate convergence to a better local maximum than that achieved by the standard EM algorithm. The adaptive data weighting strategies exploit the special characteristics of the EM algorithm. We will also show how they can be combined with the SQUAREM acceleration schemes discussed in Varadhan and Roland (Scandinavian Journal of Statistics, 2007) to obtain fast converging iterative schemes. We will evaluate the effectiveness of these strategies in two problems: a simulation example involving a finite, Gaussian mixture; and a real-world problem involving latent class modeling of the profile of multiple biomarkers in a geriatric syndrome.
|
- The address information is for the authors that have a + after their name.
- Authors who are presenting talks have a * after their name.
Back to the full JSM 2008 program |
