论文部分内容阅读
目前,学习具有隐藏变量的贝叶斯网络结构主要采用结合 EM 算法的打分-搜索方法,其效率和可靠性低.本文针对此问题建立一种新的具有隐藏变量贝叶斯网络结构学习方法.该方法首先依据变量之间基本依赖关系、基本结构和依赖分析思想进行不考虑隐藏变量的贝叶斯网络结构学习,然后利用贝叶斯网络道德图中的Cliques 发现隐藏变量的位置,最后基于依赖结构、Gibbs sampling 和 MDL 标准确定隐藏变量的取值、维数和局部结构.该方法能够避免标准 Gibbs sampling 的指数复杂性问题和现有学习方法存在的主要问题.实验结果表明,该方法能够有效进行具有隐藏变量的贝叶斯网络结构学习.
At present, the Bayesian network structure with hidden variables mainly adopts the scoring-search method combined with EM algorithm, which has low efficiency and reliability.In this paper, a new Bayesian network structure learning method with hidden variables is established in this paper. In this method, the Bayesian network structure without hidden variables is studied based on the basic dependencies, basic structure and dependency analysis, and then the position of the hidden variables is found by Cliques in the moral graph of Bayesian networks. Finally, based on the dependence Structure, Gibbs sampling and MDL standard to determine the value, dimension and local structure of hidden variables.This method can avoid the exponential complexity problem of standard Gibbs sampling and the main problems existing in existing learning methods.The experimental results show that this method can effectively Conduct Bayesian network structure learning with hidden variables.