摘要
Dirichlet过程是一种应用于非参数贝叶斯模型中的随机过程。通过其良好的聚类性质,基于此的模型可以通过简单的Gibbs采样决定参数的个数,从而为模型选择提供方便。近年来,在理论和应用上均得到了迅速的发展,引起越来越多的关注。分层Dirichlet过程是LDA模型的非参数模型推广,可以用来构建无穷个基本组元的混合模型。HDP被广泛地应用在概率话题模型的领域。首先说明Dirichlet过程的原理和采样方法,然后将其推广到分层Dirichlet过程中,并重点介绍基于Dirichlet过程的混合模型,最后对分层Dirichlet过程的应用进行了介绍。
Dirichlet process is a stochastic process widely used in non-parametric Bayesian models. Because of its good clustering properties,the models based on it can determine the number of parameters by simple Gibbs sampling,so that provides convenience for model selection. During these years,this technique has been rapidly developed in both theory and application field,and has drawn growing attentions. Hierarchical Dirichlet process is a non-parametric generalisation of Latent Dirichlet Allocation. HDP can be used to construct mixture model with infinite components and has been used widely in the domain of probabilistic topic modelling. In this paper,we first expatiate on the properties and sampling methods of Dirichlet process,and then extend it to HDP models. We thoroughly discuss the mixture models based on Dirichlet process. Finally,we introduce the applications of hierarchical Dirichlet processes.
出处
《计算机应用与软件》
CSCD
北大核心
2014年第8期1-5,41,共6页
Computer Applications and Software
基金
国家自然科学基金项目(61272539)
作者简介
周志敏，副教授，主研领域：信息系统分析与设计，分布式计算。
高申勇，博士生。
