计算机技术与发展2017，Vol.27Issue(10)：19-23,5.DOI:10.3969/j.issn.1673-629X.2017.10.005
无限狄利克雷混合模型的变分学习
Variational Learning for Infinite Dirichlet Mixture Model
摘要
Abstract
Finite Gauss mixture model is widely used in pattern recognition,machine learning and data mining and so on,but many data in reality are non Gauss,which cannot accurately describe these data. In addition,there exist difficulties in parameter estimation and model selection in the finite Gauss mixture model. In order to better fit the non Gauss data and solve the problem of parameter estimation and model selection of the finite Gauss mixture model,on the basis of research on basic learning method of infinite Dirichlet mixture model suitable for modeling the data of a non Gauss,an efficient variational approximate inference algorithm is proposed,which solves problem of parameter estimation and model selection at the same time. In order to verify its validity,a lot of experiments are carried out on the syn-thetic data set. The experimental results show it can solve the problem of model selection and parameter estimation. Infinite Dirichlet mix-ture model proposed can also be applied to object detection,text classification,image classification and other parts.关键词
狄利克雷/无限混合模型/变分贝叶斯/模型选择/参数估计Key words
Dirichlet/infinite mixture models/variational Bayes/model selection/parameter estimation分类
信息技术与安全科学引用本文复制引用
曾凡锋,陈可,王宝成,肖珂..无限狄利克雷混合模型的变分学习[J].计算机技术与发展,2017,27(10):19-23,5.基金项目
国家自然科学基金资助项目(61371142) （61371142）
北方工业大学校内专项(XN060) （XN060）
