【摘要】 本文研究在命题变量连续取值的信度网络中建立因果树模型的问题.我们指出:如果信度网络中的所有命题变量服从于某个可树分解的联合正态分布,则由网络可建立因果树.该树的拓扑结构以及树中所有非叶节点之间的概率依赖关系皆可由叶节点之间的两两相关系数确定;并且,为了由网络建立因果树,对于正态分布变量的限制要比两值分布变量的限制松.文中还给出了建立因果树的方法和步骤.更多还原

【Abstract】 This paper considers the structuring causal tree models in Belief Networks with continuous variables. It is shown that if a collection of probabilisticly coupled variables are governed by a joint normal distribution and tree structured representation exists, then both the topology and all internal relationships of the tree can be structured by observing pairwise dependencies among the available variables (i.e, the leaves of the tree). Furthermore, the constraints for normal distributed variables are less than that for bi-valued variables.更多还原