贝叶斯网络结构学习分析

Learning Bayesian Network Structure

贝叶斯网络结构学习(以下简称结构学习)的目标是寻找对先验知识和数据拟合得最好的网络结构。结构学习有两种方式,一种是模型选择,即选择一个最好的网络结构;另一种是选择性的模型平均,即选择合适数量的网络结构,以这些网络结构代表所有的网络结构。

In this paper the analysis of principle and process of Bayesian network structure learning is given. Bayesian network structure learning is a process that seeks the best network structure fitting the prior knowledge and data. The computing of posterior can be closed when data are completed and some other conditions are satisfied,while the computing is not closed when some data are missing.One solution for missing data is fill-in methods, another is to approximate the likelihood of structure, then to compute the probabilities of structure.