University of Houston

Department of Computer Science

In Partial Fulfillment of the Requirements for the Degree of

Doctor of Philosophy

Rachsuda Jiamthapthaksin

Will defend her dissertation proposal

Clustering with Plug-in Fitness Functions

for Shape Discovery

Abstract

Shapes are generic, but they carry useful information that can be used to extract knowledge from databases and images. The main goal of this dissertation is to develop methodologies for shape-aware clustering. Our research focuses on the discovery of “interesting” regions of arbitrary shape in spatial datasets in particular, and on designing shape-aware clustering algorithms in general. For this purpose, we propose clustering algorithms with plug-in fitness function. MOSAIC, an agglomerative clustering algorithm that allows for plug-in fitness functions is proposed. Specifically, the fitness function reflects the characteristics of the kind of shape the domain expert is interested in and it is plugged into the shape aware clustering algorithm. Three potential alternatives are explored to cope with shape in the clustering algorithm: creating shape signatures, learning the fitness functions, and applying density estimation techniques. Moreover, we propose the novel approach to learn fitness functions based on a domain experts’ feedback that uses regression trees and other prediction techniques. The proposed techniques will be evaluated in case studies that center on region discovery in planetary sciences, geology, and environmental sciences.

Date: Wednesday, June 6, 2007
Time: 4:00 PM
Place: 550-PGH

Faculty, students, and the general public are invited.
Advisor: Dr. Christoph F. Eick