University of Konstanz
Graduiertenkolleg / PhD Program
Computer and Information Science

Graduation Talks


Clustering in parallel universes


Bernd Wiswedel, University Konstanz
Konstanz, Germany

date & place

Wednesday, 28.06.2006, 15:15 h
Room C252


Classical data analysis methods typically assume that all objects of a data set are described in a single feature space. This feature space is assumed to comprise all necessary information to classify an object. However, in many real-world applications there are numerous ways to describe complex objects. An example are musical songs, i. e. audio streams, which can be represented based on dynamics, melody, and key or - as a different representation - based on rhythm and harmony. A third representation may be more descriptive, such as interpreter, position in music charts, length, and so on. Further examples of complex objects are images, 3D objects or molecules in drug discovery. With regard to learning, such as clustering or building classification models, it is often unclear, which of the available descriptors are optimal for any given task. Clustering in Parallel Universes is a new research field that deals with such multiply described data sets. It aims at identifying interesting patterns in data, e. g. groups of objects that cluster well in one (or few) universe(s).