University of Konstanz
Graduiertenkolleg / PhD Program
Computer and Information Science

Colloquium of the Department and the PhD Program

title

Swarm based clustering with partition validity

speaker

Prof. Lawrence O. Hall, University of South Florida
Tampa, Florida, USA

date & place

Wednesday, 20.07.2005, 16:15 h
Room Z613

abstract

Clustering algorithms may stop at local extrema. Many clustering algorithms also require some knowledge of the number of cluster centers. This talk discusses using an ant based approach to clustering. This approach can be directly distributed making it tractable for large data sets. The ants move independently in feature space with a group cooperating to form cluster centers and further participating in a set of clusters (partition). The utilization of Ants allows for a broader search of the partition space than with an iterative optimization algorithm such as fuzzy or hard c-means. Results from one artificial and six real data sets show that this approach can enable an appropriate number of cluster centers to be discovered. The final partitions are typically better optimized than partitions obtained through traditional approaches as shown in experimental comparisons on one artificial and six real data sets.