摘要
本文研究了一类基于无监督聚类学习的算法———EM算法的算法实现.EM算法通常用于存在隐含变量时的聚类学习,由于引入了隐含变量,导致算法难以保证收敛和达到极优值.本文通过将该算法应用于高斯混合模型的学习,引入重叠度分析的方法改进EM算法的约束条件,从而能够确保EM算法的正确学习.
In this paper,We conducts a theoretical analysis into the method of Machine leaning with EM algorithm which is an unsupervised-clusting one.The EM algorithm used to estimate some clusting-learning parameters including hidden variables, Which lead to difficulties of converging correctly and obtaining to the local maximum points . We use EM algorithm to learn some parameters of Gaussian Mixture Model and demonstrate that the analysis of the mixture density′s overlap measure can enforce the restrict conditions of EM algorithm, as a result ,this analysis can assure the efficiency of learning.
出处
《山西师范大学学报(自然科学版)》
2005年第1期46-49,共4页
Journal of Shanxi Normal University(Natural Science Edition)
基金
四川省教育厅重点项目基金资助(2004A102)