基于全局消息传播的贝叶斯推理

在贝叶斯网络中，常常需要作不确定概率推理。然而针对一般复杂网络，精确推理算法由于计算复杂度太高而常常被摒弃。针对这一问题，本文提出了一种基于全局传播的PPJT近似推理算法。PPJT算法采用消息传播机制，通过消息的收集与分发过程，可以更新和修正连接树节点的团势并最终生成相容连接树。与另一种常用的近似推理算法即似然权重（Likelihood Weighting）算法的时间性能对比实验显示，采用消息传播机制的PPJT算法有效地降低了计算的时间复杂度；同时与似然权重算法的性能对比实验表明，在相对小规模观察样本输入条件下，PPJT算法能够保证更高的概率推理精度。PPJT算法为实现一般复杂网络中的概率推理提供了一种新的理论工具。

Bayesian Inference Based on Global Message Propagation

Uncertain probabilistie inference is often made in Bayesian network However,for a common complicated network,accurate inference algorithm is always deserted for its unpaid high cost of computing complexity.Aiming at this problem,this paper brings forward a nearly accurate inference algorithm PPJT.Newalgorithm applies the mecha- nism of passing message to update the potentials of Join tree's cliques by steps of message collection and message dis- tribution and eventually generates a consistent join tree.Compared with another nearly accurate inference algorithm, namely likelihood weighting algorithm,the time-using performance experimentation shows that PPJT decreases the time complexity efficiently.At the same time,PPJT improves the uncertain inference accuracy.The experimentation for computing accuracy comparison shows that,under relative small samples input,PPJT can ensure much higher accu- racy for inference.PPJT provides a new theoretic tool for implementation of probabilistic inference in the common com- plicated network.