Stochastic Systems Group
Home Research Group Members Programs  
Demos Calendar Publications Mission Statement Alumni

SSG Seminar Abstract

Adaptive Clustering and Applications in Information Systems
Prof. Hava Siegelmann
Technion, Israel, and Visiting Professor, LIDS

This talk introduces adaptive techniques for data mining via clustering, and for reactive personalized searches for fast and accurate retrieval. We introduce a novel algorithm for clustering irregularly shaped data, including very close and partially overlapping clusters which are difficult to distinguish by other clustering methods. Our algorithm learns the individual "shapes" adaptively, and defines appropriate local metrics using a new notion of high-order neurons. Providing information on the clusters and their relationships, makes the algorithm ,more appropriate for datamining purposes. The identification of the data defining the clusters boundaries and distinguishing them from internal data points and from outliers can be very useful for the understanding of the clusters at hand. The support vector machine (SVM) method was recently developed by Vapnik for the classification of labeled data points by hyperplanes; we employ a similar kernel method but for unlabeled data in general rich boundaries, gaining the first and only support vector algorithm for clusters of general shapes.

Problems with this site should be emailed to