摘要
在Shenoy-Shafer赋值代数公理基础上,定义了贝叶斯网(Bayesian network,BN)推理结构上的赋值代数;提出BN上概率推理的赋值代数模型,该模型包括推理结构转变、赋值初始化、一致性消息全局传播与吸收以及概率推理的局部计算四个环节,建立了相应的算法以及联合概率与条件概率的计算方法;最后通过例子说明文章所提出的方法.所提出计算模型为贝叶斯网的概率推理提供了一种新的局部计算方法.
On the basis of Shenoy-Shafe' valuation-algebras axioms, valuation-algebras in BN (Bayesian network) inference structure are defined. A valuation-algebras model on probabilistic inference in BN is put forward, which includes four processes, namely, inference structure transformation, valuating initialization, consistency message propagation and absorption, and local computation of probabilistic inference. Corresponding algorithms are constructed and computation methods of joint probability and condition probability are concluded. The methods put forward are accounted for with an example at the end of the paper. The proposed computation models will supply new local computation methods for Bayesian network probabilistic inferences.
基金
国家自然科学基金资助项目(70471046)
教育部博士点基金资助项目(20040359004)
合肥工业大学科学研究发展基金资助项目(040301F).
关键词
赋值代数
贝叶斯网
概率推理
连接树
消息
valuation algebras
Bayesian network
probabilistic inference
junction tree
message