COMPLEXITY PURSUIT: COMBINING NONGAUSSIANITY AND AUTOCORRELATIONS FOR SIGNAL SEPARATION

Aapo Hyv#rinen, http://www.cis.hut.fi/aapo/
aapo.hyvarinen@hut.fi

A generalization of projection pursuit for time series, i.e. signals with time structure, is introduced. The goal is to ønd projections of time series that have interesting structure. We deøne the interestingness using criteria related to Kolmogorooe Complexity or coding length: Interesting signals are those that can be coded with a short code length. We derive a simple approximation of coding length that takes into account both the non­ gaussianity and the autocorrelations of the time series. Also, we derive a simple algorithm for its approximative optimization. The resulting method is closely related to blind separation of nongaussian, time­dependent source signals.