Online Program

Return to main conference page
Saturday, June 1
Computational Statistics
The IMS Program on Self-Consistency: a Fundamental Statistical Principle for Deriving Computational Algorithims
Sat, Jun 1, 1:00 PM - 2:35 PM
Grand Ballroom J

Latent Variable Models, Self-Consistency, and Stochastic Approximation (305059)

Zexi Song, Rutgers University 
*Zhiqiang Tan, Rutgers University 

Keywords: latent variable, self consistency, stochastic approximation

Latent variable models are widely used, such as generalized linear mixed models and others. For such models, a general class of estimators can be defined by self-consistency, which includes the maximum likelihood estimator in the special case where a parametric likelihood function is available. We show that such estimators can be efficiently computed by stochastic approximation algorithms, along with the estimated variances.