摘要
提出了一种基于混合团树的智能推理体系架构,利用原始贝叶斯网络中变量之间的依赖关系对连接树进行改造,使得在推理过程中满足以下两个优势:(1)它能直接消除与证据和查询的无关的变量集,使得混合团树得以缩减为更小规模的二级树状结构,进而使消息不必在所有节点中传播;(2)它继承了连接树传播算法中可以重复利用混合团树中预先储存的信息进行加速推理.
An intelligent reasoning architecture based on mixed clique trees is proposed.It uses the dependence between variables in the original Bayesian network to transform the join tree,so that the following two advantages are satisfied in the reasoning process:(1)It can directly eliminate the variable set unrelated to evidence and query variables,so that the mixed clique tree can be reduced to a smaller secondary tree structure,and thus messages do not need to be propagated in all nodes.(2)It inherits advantages of the join tree propagation algorithm,that is it can reuse the pre-stored information in the mixed clique trees so as to accelerate reasoning.
作者
郑靓
孙毅
ZHENG Liang;SUN Yi(College of Mathematics and System Science,Xinjiang University,Urumqi 830017,China)
出处
《东北师大学报(自然科学版)》
CAS
北大核心
2023年第2期35-44,共10页
Journal of Northeast Normal University(Natural Science Edition)
基金
国家自然科学基金资助项目(11726630,11701491)。
关键词
贝叶斯网络
智能推理
连接树传播
混合团树
Barren团
Bayesian network
intelligent reasoning
join tree propagation
mixed clique tree
Barren clique