GraphMaxEnt
Assume a undirected graph G on a finite domain X and a probability distribution P on X. Graph entropy, defined in terms of the vertex packing polytope, is recast as , with suitably defined plausibilit...

Toward a Generalized Bayesian Network
The author's past work in this area has shown that the probability of a state of a Bayesian network, found using the standard Bayesian techniques, could be equated to the Maximum Entropy solution and ...

AIP Conf. Proc. / Volume 872 / Issue 1

Online Learning in Discrete Hidden Markov Models

AIP Conf. Proc. -- November 29, 2006 -- Volume 872, pp. 187-194
Bayesian Inference and Maximum Entropy Methods In Science and Engineering; doi:10.1063/1.2423274

Issue Date: 29 November 2006

You are not logged in. Log in

ABSTRACT

Roberto Alamino^* and Nestor Caticha
^*Neural Computing Research Group, Aston University, Aston Triangle, Birmingham, B4 7ET, United Kingdom
Instituto de Física, Universidade de São Paulo, CP 66318 São Paulo, SP, CEP 05389-970 Brazil

Wepresent and analyze three different online algorithms for learning indiscrete Hidden Markov Models (HMMs) and compare their performance withthe Baldi-Chauvin Algorithm. Using the Kullback-Leibler divergence as a measureof the generalization error we draw learning curves in simplifiedsituations and compare the results. The performance for learning driftingconcepts of one of the presented algorithms is analyzed andcompared with the Baldi-Chauvin algorithm in the same situations. Abrief discussion about learning and symmetry breaking based on ourresults is also presented. ©2006 American Institute of Physics

Permalink:

http://link.aip.org/link/?APCPCS/872/187/1

BUY THIS ARTICLE

(US$28)