基于贝努里分布的贝叶斯网络结构学习算法  

Learning Bayesian Network Structure Based on Bernoulli Distribution

作　　者：孙岩[1] 吕世聘[2] 唐一源[3] 

机构地区：[1]辽宁师范大学计算机与信息技术学院 [2]大连理工大学计算机科学与工程系,大连116023 [3]大连理工大学神经信息学研究所,大连116023

出　　处：《计算机科学》2008年第1期168-170,共3页Computer Science

基　　金：国家自然基金项目(60472017,30670699)资助课题

摘　　要：目前,学习具有丢失数据的贝叶斯网络结构主要采用结合EM算法的打分-搜索方法和基于依赖分析的思想,其效率和可靠性比较低。本文针对此问题建立一个新的具有丢失数据的贝叶斯网络结构学习算法。该方法首先根据贝努里分布来表示数据库中变量结点之间的关系,并用Kullback-Leibler(KL)散度来表示同一结点的各个案例之间的相似程度,然后根据Gibbs取样来得出丢失数据的取值。最后,用启发式搜索完成贝叶斯网络结构的学习。该方法能够有效避免标准Gibbs取样的指数复杂性问题和现有学习方法存在的主要问题。At present, the algorithm of learning bayesian structure with missing data is mainly based on the search and scoring method combined with EM algorithm. The algorithm has low efficiency. In this paper , a new algorithm of learning Bayesian network structure with missing data is presented. First, we adopt the Bernoulli distribution to express the relationship between the variables in database. Second, we use KL divergence to express the similarity between the cases. Third, we draw the value of the missing data according to the Gibbs sampling. Finally, we use heuristical search to complete the learning of Bayesian network structure. This method can avoid the exponential complexity of standard Gibbs sampling and the main problems in the existing algorithm.

关 键 词：贝努里分布 KL散度 贝叶斯网络 Gibbs取样 

分 类 号：TP18[自动化与计算机技术—控制理论与控制工程]