Vorträge

Linear and Nonlinear Methods for Decomposing (Brain)Signals

Prof. Dr. Klaus-Robert Mueller
Fraunhofer FIRST.IDA and University of Potsdam

I will first review the technique of linear independent component analysis (ICA). Then I report on a study where we applied ICA to process MEG recordings of near DC-fields over the auditory cortex.

The extraction of DC fields from MEG recordings is highly interesting for medical applications, since slowly varying DC-phenomena have been found e.g. in anoxia and spreading depression in animals and they also play an important role in diagnosis of stroke and ischemic diseases.

Comparing several blind decomposition approaches, it turns out that only our TDSEP algorithm successfully extracts a component that can be interpreted as an acoustical "sustained field". The task is challenging because of the limited amount of available (high dimensional) data and the corruption by outliers, therefore we propose this application as a real-world testbed for studying the robustness of ICA methods.

If time permits we will enter the realm of non-linear ICA and use kernel-based learning methods to provide an elegant solution to the nonlinear demixing problem.

Joint work with Wubbeler, G., Ziehe, A., G., Mackert, B.-M., Trahms, L., Curio, G., Kawanabe, M., Harmeling, S. and Blankertz, B.

Zeit: Donnerstag, den 11. Juli 2002, 16 Uhr c.t.

Ort: Institut für Neuro- und Bioinformatik
        TZL, Seelandstr. 1a, Geb. 5
        Seminarraum OG.3