【摘要】 <正> 信度网提供了一套强有力的图形工具来表达基于概率的领域知识,并且已经成功地应用在诸多实际问题中,如:故障诊断、预测、模式识别、数据挖掘等。目前基于信度网提出了多种推理算法以精确计算待求概率值。这些算法在某些信度网上具有很高的推理效率,但是在最坏情况下这些算法的计算时更多还原

【Abstract】 This paper introduces two principal approximate inference methods on belif network, the method based on search techniques and the method based on simulation techniques. The basic idea of the search method and the bounded conditioning algorithm used in approximate inference are presented in the introduction of the first method. During the introduction of the second method, the basic idea of Monte Carlo method is presented first, then the paper focuses on three principal algorithms of that method,which are straight simulation algorithm,likelihood weighting algorithm and bounded variance algorithm ,and on several principal sampling techniques,which involve reject sampling, importance sampling, forward sampling and Gibbs sampling. The current trend of the research of this area is presented at the end of this paper.更多还原