摘要
EM算法是一种应用十分广泛的算法,对含有隐变量的概率模型参数的估计往往非常有效。在实际应用EM算法时,隐变量如同参数初值一样可以有多种选择或构造方式,而隐变量对EM算法的影响却少被关注。本文通过一个具体的离散概率模型参数的估计探讨了隐变量的选择对EM算法的影响,构造了两种不同的隐变量,导出了两种不同的EM迭代格式,并用牛顿法给出了常规的迭代格式,对它们进行了对比。数据实验表明:隐变量选择的不同不影响EM算法的收敛性,但隐变量的选择对EM算法收敛速度的影响却很大。
EM algorithm is a widely used algorithm.It is very effective to estimate the parameters of probability models with hidden variables.In the practical application of EM algorithm,the hidden variables can be selected or constructed in many ways just like the initial values of parameters,however,little attention has been paid to the influence of hidden variables on EM algorithm.In this paper,the influence of the choice of hidden variables on EM algorithm is discussed by estimating the parameters of a discrete probability model.Two different hidden variables are constructed and two different EM iterative schemes are derived,the conventional iterative scheme is given by Newton method.Then these iterative schemes are compared.Data experiments show that the choice of hidden variables does not affect the convergence of EM algorithm,furthermore,the choice of hidden variables has a great influence on the convergence speed of EM algorithm.
作者
刘芝秀
吕凤姣
李运通
LIU Zhi-xiu;LV Feng-jiao;LI Yun-tong(Department of Science,Nanchang Institute of Technology,Nanchang 330099,China;Engineering Department,Huanghe Science and Technology College,Zhengzhou 450063,China;Department of Basic Course,Shaanxi Railway Institute,Weinan 714025,China)
出处
《安徽师范大学学报(自然科学版)》
2022年第3期221-226,共6页
Journal of Anhui Normal University(Natural Science)
基金
江西省教育厅科技项目(GJJ190963)
陕西铁路工程职业技术学院基金项目(KY2019-46).
关键词
EM算法
隐变量
收敛速度
参数估计
EM algorithm
hidden variables
convergence rate
parameter estimation