SEMINAR
Room 27, Statistical Laboratory
University of Cambridge
*********************************
Friday, 23 April, 1999
*********************************
2.05 Arnaud Doucet (Engineering Department, Cambridge)
SOME MCMC ALGORITHMS FOR BAYESIAN PARAMETER ESTIMATION OF HMM
Hidden Markov models (HMM) have numerous applications in statistical
signal processing and applied statistics (speech processing, target
tracking etc.).
The standard method to perform parameter estimation is the EM algorithm.
This deterministic algorithm suffers from several drawbacks. That is why
several stochastic versions of the EM as well as Markov chain Monte Carlo
(MCMC) methods have been developed.
In this talk, I will present some original MCMC-based algorithms to
perform maximum a posteriori parameter estimation of HMM. These
algorithms are simulated annealing type algorithms that make the most of
the statistical structure of the model and
can also be applied to more general problems involving missing data.
A few applications will be presented.
A part of this work has been realized in collaboration with C. Andrieu,
S.J. Godsill (Dept. Engineering, Cambridge) and C.P. Robert (Statistics
Lab., CREST, INSEE, Paris).
*********************************
ALL INTERESTED ARE WELCOME
Susan Pitts
Organizer,
Statistical Laboratory Seminars.
UNIVERSITY OF CAMBRIDGE
DEPARTMENT OF PURE MATHEMATICS AND MATHEMATICAL STATISTICS
STATISTICAL LABORATORY
16 MILL LANE, CAMBRIDGE CB2 1SB
Tel: (01223) 337958
Fax: (01223) 337956
%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%
|