摘要
在本文中，直接利用计算概率分布的办法证明了在只包含离散变量的反馈系统产生的因果图中的条件独立关系可以由一分离识别出．
It is well known that d-separation can identify conditional independence in Bayesian network. Pearl J, Dechter R~[6] proved d-separation criteria is sound in causal graphs with feedback assuming V is discrete variable set and the state space of V is finite. In this paper it is shown that their conclusion is valid only under certain constraints on the causal equations and under the same constraints, their conclusion is generalized by computing the probability distriubution to the situation in which the range of V is in the discrete but infinite state space. Counterexamples show when the above constraints does not hold or V is continuous, the d-separation criteria is not valid any more.
出处
《应用数学学报》
CSCD
北大核心
2000年第2期299-310,共12页
Acta Mathematicae Applicatae Sinica
基金
国家自然科学基金基金!(# 19871003
#39930160)
博士点基金!(#97000139)
