UNIVERSITY OF COLORADO AT DENVER

PLACE: Mathematics Conference Room 626 UCD Building, 1250 14th St., Denver

TIME: NOON (Refreshments served at 11:45 am)

DATE: April 16, 2001

Title:
 Optimization of Restricted Likelihood Functions for Covariance
       Analysis of Mixed Model Equations


Speaker:
 Arnold Neumaier, University of Vienna, Austria



Abstract:
 In animal breeding and in labor market analysis, huge amounts of
data are available that need to be interpreted in terms of so-called mixed
model equations, a clan of Bayesian stochastic models that allows to infer
unmeasured properties of individuals such as the additive genetic value of
animals, or the weight with which certain economical factors affect a
person's income.

If the covariance structure of mixed model equations is known, a large
sparce least squares problem must be solved.  However, the estimation of
the covariance structure requires the maximization of the restricted
likelihood (REML) which is a huge optimization problem.

In this talk I'll report on techniques that allowed the successful
solution of these optimization problems for animal breeding applications
with 250,000 variables and up to 50 covariance parameters, and on current
attempts to scale up the technique to work for 10-100 times bigger
problems from the analysis of labor data from France and the USA.