|
Activity Number:
|
560
|
|
Type:
|
Contributed
|
|
Date/Time:
|
Thursday, August 6, 2009 : 8:30 AM to 10:20 AM
|
|
Sponsor:
|
Social Statistics Section
|
| Abstract - #305108 |
|
Title:
|
Line Search Algorithm for Maximum Likelihood in Exponential Families
|
|
Author(s):
|
Saisuke Okabayashi*+ and Charles J. Geyer
|
|
Companies:
|
The University of Minnesota and The University of Minnesota
|
|
Address:
|
313 Ford Hall, Minneapolis, MN, 55455,
|
|
Keywords:
|
Markov chain Monte Carlo ; exponential families ; maximum likelihood ; social networks ; exponential random graph models
|
|
Abstract:
|
Maximum likelihood estimation for exponential families can be difficult when the log likelihood is expensive to compute. Often, especially when the log likelihood is evaluated by Markov chain Monte Carlo, it is easier to evaluate the gradient of the log likelihood (observed minus expected value of the natural statistic) than to evaluate the log likelihood itself. We present a line search algorithm that converges to the maximum likelihood estimate (MLE) of a full exponential family when the MLE exists and is unique. Unlike other algorithms, this algorithm utilizes first derivative information only, evaluating neither the log likelihood function itself nor derivatives of higher order than first.
|
