Bereiche | Tage | Auswahl | Suche | Downloads | Hilfe
DY: Dynamik und Statistische Physik
DY 21: Nichtlineare Stochastische Systeme
DY 21.5: Vortrag
Dienstag, 12. März 2002, 11:00–11:15, H3
A stochastical method to find independent components using temporal information — •Andreas Jung1 and Andreas Kaiser2 — 1Institut für Theoretische Physik, Universität Regensburg — 2Max Plank Institut für Physik komplexer Systeme, Dresden
Recently a newly developed (non-)linear stochastical method for data analysis has gained wide-spread attention, the independent component analysis (ICA). Goal of this method is to find a transformation from the observed signals to some new representation, so that the new components have minimal stochastical dependence. Such a representation seems to capture the essential structure of the data in many applications.
In classical ICA-Algorithms the dynamics is not taken into account; we will therefore present an extension of this method, using also temporal information to enhance the separation quality of the independent components, since in many real world problems, the data sets have time structure information. Furthermore, we are also able to separate Gaussian sources and reduce the effect of noise under some circumstances.