Andreas Galka
(Institute for Medical Psychology and Medical Sociology, University of Kiel)
Title
Blind Signal Separation by Linear Gaussian Modelling: The EM Algorithm Revisited
Abstract
The analysis of biomedical time series represents a field of intense and challenging research.
This talk deals with linear Gaussian state space modelling and Kalman filtering, representing a very general class of algorithms for the modelling of time series.
In particular, the application to Blind Signal Separation will be discussed.
Practical results will be shown for an extensive simulation and for several real data sets from clinical practice.
The specific weaknesses of Independent Component Analysis (ICA) will be demonstrated.
Finally new results regarding Expectation Maximisation (EM) as an algorithm for estimating the parameters in certain structured linear Gaussian state space modells will be presented.
