论文部分内容阅读
影响图模型选择中存在数据依赖性、计算复杂性和非概率关系问题.通过对影响图结构进行分解,提出PS-EM 算法对影响图的概率结构部分进行模型选择.给出一种 BP 神经网络,通过对局部效用函数的学习实现效用结构部分的模型选择,并引入权重阈值来避免过拟合.PS-EM 算法是在 SEM 算法中引入一种融合先验知识的MDL 评分标准来降低传统 MDL 评分对数据的依赖性,并通过将参数学习和结构评分分开计算提高计算效率.算法比较的结果显示 PS-EM 比标准 SEM 的时间性能好、对数据依赖性小,且效用部分的结构选择易于实现.
There are data dependence, computational complexity and non-probabilistic relationship problems in the influence graph model selection.By decomposing the influence graph structure, the PS-EM algorithm is proposed to select the probabilistic structure part of the influence graph.A BP neural network , Through the study of the local utility function to realize the model selection of the utility structure part and introduce the weight threshold to avoid over fitting.PS-EM algorithm is to introduce a fusion MDL scoring standard of prior knowledge into the SEM algorithm to reduce the traditional MDL The dependence of the score on the data and improve the computational efficiency by separating the parameter learning and the structural score separately.The results of the algorithm comparison show that the PS-EM has better time performance than the standard SEM, less dependence on the data, and the structure selection of the utility part is easy achieve.