利用互信息学习贝叶斯网络结构  

Learning Bayesian network structures based on mutual information

作　　者：李冰寒[1] 高晓利[1] 刘三阳[1] 李战国[2] 

机构地区：[1]西安电子科技大学数学系,陕西西安710071 [2]西安交通大学机械工程学院,陕西西安710049

出　　处：《智能系统学报》2011年第1期68-72,共5页CAAI Transactions on Intelligent Systems

基　　金：国家自然科学基金资助项目(60974082)

摘　　要：由数据构造贝叶斯网络结构是NP-难问题,因此提出了一种基于互信息的改进算法.该算法根据互信息构造初始框架,其次利用最大支撑树算法精简初始框架,并通过条件独立测试添加方向,最后利用贪婪算法得到最优网络结构.数值实验表明,改进算法无论是在B IC的得分值,还是在结构的误差上都有一定的改善,并且在迭代次数、运行时间上均有明显降低,能较快地确定出与数据匹配程度最高的网络结构.Constructing Bayesian network structures from data is an NP-hard problem, and an improved algorithm was proposed based on mutual information. This algorithm built the initial skeleton using mutual information, refined the initial skeleton by employing the maximum spanning tree algorithm, and then oriented edges according to conditional independence tests. Finally, the optimal network structure was obtained using a greedy search. Numerical experiments show that both the BIC score and structural error made some improvements from previous results, and the number of iterations and running time was greatly reduced. Therefore the structure with highest degree of data matching was shown to be relatively faster as determined by the improved algorithm.

关 键 词：贝叶斯网络 结构学习 互信息 独立测试 最大支撑树 

分 类 号：TP181[自动化与计算机技术—控制理论与控制工程]