修正Gibbs采样的有限混合模型无监督学习算法

被引:3
作者
刘伟峰
韩崇昭
石勇
机构
[1] 西安交通大学电子与信息工程学院
关键词
无监督学习; 有限混合模型; 参数维数变化; 跳变; 分布元管理;
D O I
暂无
中图分类号
TP391.41 [];
学科分类号
080203 ;
摘要
针对传统有限混合模型无监督学习算法不能处理参数维数变化的问题,提出了一种基于修正Gibbs采样的无监督学习算法.该算法的关键是,在每一次完全采样之后引入分布元的合并和剔除技术,即将利用均值、协方差矩阵差值的2范数作为合并的判断准则,最小且小于阈值的分布元权重作为剔除规则.仿真实验表明,所提算法对于参数初值的选择是不敏感的,对于分布元个数的先验信息要求得更少,它不仅可以处理维数变化问题,而且不必计算跳变概率,同时能够很好地估计出分布元个数及其参数.
引用
收藏
页码:15 / 19
页数:5
相关论文
共 10 条
[1]  
Unsupervised learning of finite mixture models. Figueiredo M A T and Jain A K. IEEE Transactions on Pattern Analysis and Machine Intelligence . 2002
[2]  
Markov chain monte-carlo estimation of classical and dynamic switching and mixture models. Fruhwirth-Schnatter,S. Journal of the American Statistical Association . 2001
[3]  
Reversible jump Markov chain MonteCarlo computation and Bayesian model determination. GREEN P J. Biometrika . 1997
[4]  
On Bayesian analysis of mixtures with an unknown number of components. Richardson,S.,Green,P. Journal of Royal Statistical Society B . 1997
[5]  
TRUST-TECH-based expectation maximiza-tion for learning finite mixture models. REDDY C K,CHIANG Hsiao-Dong,RAJARAT-NAM B. IEEE TransPatt Anal Mach Intell . 2008
[6]  
Finite mixture models. McLachlan G and Peel D. . 2000
[7]  
Estimation of finite mixture distributions through Bayesian sampling. Diebolt,J.,Robert,C.P. Journal of the Royal Statistical Society B . 1994
[8]  
Feature and model selection forGaussian mixture models. CONSTANTINOPOULOS C,,TITSIAS M K,BAYESIAN A L. IEEE Trans Patt AnalMach Intell . 2006
[9]  
A hybrid Sem algorithm forhigh-dimensional unsupervised learning using a finitegeneralized dirichlet mixture. BOUGUILA N,ZIOU D. IEEE Transactions on Image Processing . 2006
[10]  
Genetic-based EM algorithm for learning Gaussian mixture models. Pernkopf,F,Bouchaffra,D. IEEE Transactions on Pattern Analysis and Machine Intelligence . 2005