摘要
删除Bayes网络中的弧以减小网络结构的复杂性 ,从而降低概率推理算法的复杂度是一种对Bayes网络进行近似的方法 .该文讨论了在删除Bayes网络中的一条弧之后得到的最优近似概率分布和原概率分布之间的关系 ,证明了对满足一定条件的结点子集而言 ,其边缘概率分布在近似以后具有不变性 .
As a model simplification method, arc removal is a way to approximate Bayesian networks by introducing additional conditional independency to simplify the network structures. Since the simplified network is no longer a I-map of the joint probability distribution, the best we can do is to approximate the probability distribution as close as possible under some criteria, such as minimal Kullback-Leibler divergence. Although the joint probability distribution of the whole random variables will be changed after we remove an arc, it is possible that the marginal probability distribution on some subset of the variables will not be changed. This paper proves the invariance of the marginal probability distribution on some set of the variables in the optimal approximate Bayesian networks.
出处
《计算机学报》
EI
CSCD
北大核心
2004年第7期993-997,共5页
Chinese Journal of Computers
基金
国家自然科学基金 ( 60 0 73 0 5 3 )资助
关键词
BAYES网络
d-分割
弧的删除
边缘概率分布
Bayesian networks
d-separation
arc removal
marginal probability distribution