Date: 2006/jan/16 14:00-
Invited speaker: Professor Andreas Galka
venue: Dw601, Institute of Industrial Science, The University of Tokyo

title: EEG Time Series Decomposition by State Space Modelling

abstract:
Independent Component Analysis (ICA) has become well-known as a set of
techniques for decomposing multivariate time series into independent
source components. In this talk I will compare the performance of ICA
algorithms with an alternative approach based on more classical
concepts from time series analysis. The key element will be the
estimation of state space models. Unlike most ICA algorithms, such
algorithms are dynamic, i.e., they take the time ordering of the data
into account. Using several examples from electroencephalography, it
will be demonstrated that superior decompositions can be obtained by
time series analysis, albeit at the price of higher computational
expense.