Title: Active Information Systems, Advanced Clustering, and Bio Informatics

Abstract: With the growing amount of data in information systems, it became crucial to extract information with high precision and with accurate ranking. We describe an automatic interactive method that guides the user to reach the most relevant data. This algorithm can be applied on the top of any search engines or decision support systems. [Joint work with Tommi Jaakkola, MIT]

We will introduce two clustering algorithms for points in metric spaces that allow for highly irregular shapes. Both methods are particularly applicable for data mining in the sense by reporting the geometric characteristics of the clusters. One is based on tensor multiplication and Hebb rule of unsupervised learning. It allows for both very close, non-convex, and overlapping clusters and reports geometric features of high order statistics. [join work with Hod Lipson, Cornell] The second is a hierarchical methods based on kernel functions; it reports the support vectors of the clusters boundaries. [join work with David Horn, Asa Ben-Hur and Vladimir Vapnik]

Some current work on BioInformatics with colleagues at UMass Amherst will be described as well.