论文部分内容阅读
我们经常需要修改网络参数以在局部信任度改变时集成新的概率信息,贝叶斯网络的敏感性分析就是研究网络参数与输出概率之间的关系.首先剪枝掉那些不能到达证据的节点,然后采用联合树算法对剪枝后的贝叶斯网络进行的敏感性分析,该算法在每一簇中缓存中间结果以加速计算P(e)相对于下一变量P(xi|ui)的一阶偏导;并且能够在各个簇中并行地计算导数.我们利用偏导的计算结果对两类网络参数进行敏感性分析:调整单个参数和调整整个条件概率表,并分别提出算法来进行处理.最后,实验结果证实了我们的算法在进行敏感性分析时的高效性.
We often need to modify the network parameters to integrate new probability information when local trust changes.The sensitivity analysis of Bayesian network is to study the relationship between network parameters and output probability.First, we cut out those nodes that can not reach the evidence, Sensitivity analysis of the pruned Bayesian network is then performed using a joint tree algorithm that caches intermediate results in each cluster to speed up the computation of P (e) relative to the next variable P (xi | ui) Order partial derivatives, and the derivatives can be calculated in parallel in each cluster.We conduct the sensitivity analysis of two types of network parameters by using the partial derivative results: adjusting individual parameters and adjusting the entire conditional probability table, and then propose algorithms to deal with them separately. Finally, the experimental results confirm the efficiency of our algorithm in sensitivity analysis.