Seminars

Maximizing Nondeterministic Likelihoods by Directed Stochastic Searching Algorithm

70
reads

Sheng-Mao Chang

2010-09-24
12:30:00 - 14:30:00

405 , Mathematics Research Center Building (ori. New Math. Bldg.)

In this article, we propose a directed stochastic searching algorithm which is defined as a root or optimal parameter searching algorithm where the searching directions are random. This method is useful when the objective function is complex or nondeterministic, having no closed-form representation. When the solution is unique the distance process induced from the algorithm is shown to be ergodic. The ergodic mean is therefore a measure of the precision of the root or parameter estimate. We demonstrate the usefulness of the proposed algorithms for finding the maximum likelihood estimates when the corresponding likelihood is deterministic or nondeterministic both by simulations and real cases. Finally, we comment on the limitation of DSSA and then relate DSSA to the MM algorithm and discuss the consequence when multiple solutions exist.