Discrete and Continuous Dynamical Systems - Series B (DCDS-B)

Bayesian online algorithms for learning in discrete hidden Markov models

Pages: 1 - 10, Volume 9, Issue 1, January 2008      doi:10.3934/dcdsb.2008.9.1

       Abstract        Full Text (255.6K)       Related Articles

Roberto C. Alamino - Neural Computing Research Group, Aston University, Main Building, Birmingham, B7 4ET, United Kingdom (email)
Nestor Caticha - Instituto de Física, Universidade de São Paulo, CP 66318, São Paulo, SP, CEP 05389-970, Brazil (email)

Abstract: We propose and analyze two different Bayesian online algorithms for learning in discrete Hidden Markov Models and compare their performance with the already known Baldi-Chauvin Algorithm. Using the Kullback-Leibler divergence as a measure of generalization we draw learning curves in simplified situations for these algorithms and compare their performances.

Keywords:  HMM, online algorithm, generalization error, Bayesian algorithm.
Mathematics Subject Classification:  Primary: 68T05; Secondary: 60J20, 62F15.

Received: August 2006;      Revised: September 2007;      Available Online: October 2007.