Seminar on Computational Learning and Adaptation


 
Using Background Knowledge to Constrain Clustering

Kiri Wagstaff
Department of Computer Science, Cornell University, and
DaimlerChrysler Research and Technology North America
wkiri@cs.cornell.edu

Clustering algorithms are designed to automatically uncover interesting structure in data. However, this unsupervised approach can be profitably augmented by the addition of background information. In this talk, I will discuss existing methods for using knowledge to constrain clustering and highlight their limitations. I will also present a new method that overcomes those limitations, as well as results that demonstrate clear improvements in learning ability. Finally, I will cover three proposed application areas that will be used to develop and evaluate other constraint-capable clustering algorithms.
Some of this material is available, in a written thesis proposal, at http://www.cs.cornell.edu/home/wkiri/research/wagstaff-proposal.ps


Date: Thurs., June 22

Time: 4:15-5:30PM

Place: Cordura 100


Return to the seminar schedule