Colloquium Computing Science



Date:                            Monday, March 3 2008

Speaker:                      Thomas Villmann, University Leipzig
                                    Medical Department / Clinic for Psychotherapy
                                    Computational Intelligence Group

Room and time:           5161.0267 (Bernoulliborg), 16.00




Title:    Fuzzy Classification and Class-Imaging Using Fuzzy-labelled Self-
Organizing Maps





In this talk we present an extension of the well-known self-organizing map
(SOM) for fuzzy classification. Originally, SOMs are unsupervised vector
quantizers which map high-dimensional vectorial data onto a low-
dimensional grid in a topology preserving manner under certain conditions.
We extend this approach to a supervised learning scheme for fuzzy
classification. Moreover, this supervised version gives the ability to detect
class similarities. The visualization possibilities of the usual SOM can also
be extended in this supervised variant for class imaging and, therefore, for
visualization of classification results. Thereby, the topology preserving map
property is used. We demonstrate the power of the method for two
exemplary applications: the classification non-neoplastic cell populations in
breast cancer tissue sections (mass-spectrometry data) and classification of
ground covers in satellite remote sensing.




Colloquiumcoördinator is Prof.dr. M. Aiello, Dr. M. Biehl (e-mail: