next up previous
Next: Introduction Up: Workpage of Harri Valpola

Fast Algorithms for Bayesian Independent Component Analysis

Harri Valpola and Petteri Pajunen

Helsinki University of Technology, Neural Networks Research Centre
P.O.Box 5400, FIN-02015 HUT, Espoo, Finland


Fast algorithms for linear blind source separation are developed. The fast convergence is first derived from low-noise approximation of the EM-algorithm given in [2], to which a modification is made that leads as a special case to the FastICA algorithm [5]. The modification is given a general interpretation and is applied to Bayesian blind source separation of noisy signals.


Harri Valpola