讲座题目：Test for the order of finite mixture models

主  讲  人：加拿大不列颠哥伦比亚大学陈家骅教授

时        间：2018年10月22日（周一）13：30-14：30

地        点：6号学院楼415教室

主办单位：002cc白菜资讯

摘  要：

The problem of testing for the order of finite mixture models has a long history and remains a hot research topic for a foreseeable future. Since Ghosh (1985) pointed out the undesirable asymptotic properties of the likelihood ratio test in this context, there have been marked progresses toward addressing this challenging problem. The seemingly neat and well-defined problem is challenging numerically, mathematically and statistically. The most proud development of my research team have been the modified likelihood ratio test and the EM-test. The idea of EM-test has been extensively explored for independent and identically distributed observations, for data from hidden Markov model, and for data from regression mixtures. It possesses neat solutions to finite mixture of univariate normal distributions and several other mixtures. At the same time, it seems to be less powerful when handling mixtures of general multi-parameter distributions. In this talk, we present our recent work on finite mixture of distributions from location-scale families. We show that the EM-test has manageable limiting distributions in general, and neat chi-square limiting distributions for some important examples.

主讲人简介：

    陈家骅教授，加拿大英属哥伦比亚大学国家一级科研讲座教授，国际著名统计学家。曾任泛华统计学会主席、加拿大统计学杂志主编等职务。 陈家骅教授研究兴趣广泛，在混合模型、试验设计、经验似然、大样本理论和变量选择等研究领域有突出的科研成果，在Annals of Statistics, Journal of the American Statistical Association, Biometrika, Journal of the Royal Statistical Society (series B) 等国际顶尖统计杂志上发表了学术论文100余篇。曾获得加拿大统计学会金奖、泛华统计学会杰出成就奖等荣誉。