PhD Qualifying Examination


"Density and Graph Based Clustering Algorithms"

Mr. Kai Zhang


Abstract:

Clustering is a basic and important problem in pattern recognition and
machine learning, and has received numerous attention in a variety of
application fields. In this paper, we make a brief review on two kinds of
clustering paradigms, density based and graph based clustering approaches.
The former one focuses on seeking a reliable model of the underlying data
distribution, such as the maximum likelihood inference of Gaussian mixture
models (parametric), and the mean shift clustering (nonparametric). The
latter one, in particular, the spectral clustering method, uses the
eigenvectors of the affinity matrix that encodes the local data structure
to make a global decision of clustering. These two kinds of methods have
been successfully applied in the fields of computer vision, medical
imaging, machine learning, data mining, and design automation. We
introduce their basic conceptions, related algorithms, and a number of
applications. At last we identify some connections between these two kinds
of clustering approaches, which provide directions for our future work.


Date:     		Monday, 19 December 2005

Time:                   10:00a.m.-12:00noon

Venue:                  Room 1504
			lifts 25-26

Committee Members:      Dr. James Kwok (Supervisor)
			Dr. Qiang Yang (Chairperson)
			Dr. Albert Chung
			Dr. Dit-Yan Yeung


**** ALL are Welcome ****