Learning Hidden Variables in Bayesian Network Based on Explanation Ability
Liu Feng-xia · Journal of Chinese Computer Systems · 2009
At present,the purpose of inserting hidden variables in Bayesian network is to improve the efficiency of reasoning by simplify structure.But the reliability of reasoning may fall because of adding hidden variables irrelevantly.In this paper,the local most probable explanation ability is adopted as the criterion of inserting hidden variables.And a star structure and Gibbs sampling is combined with most probable explanation to make sure the the values and optimal dimension of hidden variables.Therefor,the efficiency and reliability can be improved by inserting hidden variables in this manner.