学术报告报告题目：High dimensional clustering：A two-step method for mixture data
报 告 人：刘一鸣 博士(新加坡南洋理工大学)
报告时间：2019年9月30日（星期一）下午 15:00-16:00
报告地点：4号楼318室
邀 请 人：王绍臣 博士
欢迎广大师生前往！
数学学院
2019年9月29日
报告摘要：
Clustering is an important subject in unsupervised learning. It is a common technique used in many fields, including machine learning, statistics, bioinformatics, and computer graphics. To classify different samples into a homogeneous group, it is based on different criterions. In this talk, we focus on the clusters that are characterized by the different parameters of means and covariances, and we study the clustering method for the high dimensional mixtures. According to this setting, we propose a new method, Two-step method, to conduct clustering. Two-step method is investigated from two aspects, i.e., covariances and means, and based on the random matrix theory. Both theoretical and numerical properties of the Two-step method are discussed. Specifically, we propose two separate algorithms and one universal algorithm that are applicable to do the clustering in different settings. In addition, we prove that the misclustering error for all these three algorithms converges to zero with probability tends to one under mild conditions. Simulation studies also demonstrate that the Two-step method outperforms other methods under variety of settings.