报告人:张 寅 (香港中文大学(深圳))

日  期:2019年3月28日

时  间：14:30

地  点:理科楼 LD402

摘  要: Data clustering is a fundamental unsupervised learning strategy. The most popular clustering method is arguably the K-means algorithm, albeit it is usually applied not directly to a dataset given, but to its image in a processed "feature space" (such as in spectral clustering).  The K-means algorithm, however, has been observed to have a deteriorating performance as K, the number of clusters, increases. In this talk, we will examine some clustering models from a subspace-matching viewpoint and promote a so-called K-indicators model.  We apply a partial convex-relaxation scheme to this model and construct "essentially deterministic" algorithms that require no randomized initialization.  We will present theoretical results to justify the proposed algorithms and give extensive numerical results to show their superior scalability over K-means as the number of clusters grows.

(Collaborators: Feiyu Chen, Yuchen Yang, Liwei Xu and Taiping Zhang)

报告人简介:张寅,现为香港中文大学（深圳）校长讲座教授，并担任数据与运筹科学研究院共同院长。同时也是美国莱斯大学计算与应用数学系终身正教授。张寅教授的主要研究领域为最优化算法设计、分析、实现，以及各类实际应用和相应的计算机软件开发。张寅教授在国际顶尖的同行评审学术杂志上发表过八十多篇学术论文，在国内国外多次获得相关学会协会和杂志颁发的最佳论文奖，并且在国际国内学术会议上或学术研究机构中应邀进行过上百次报告。

学院联系人:曾 芳

欢迎广大师生积极参与！