Advances in Independent Component Analysis and Learning Machines

0444036 - ÚTIA 2016 RIV GB eng M - Část monografie knihy
Koldovský, Zbyněk - Tichavský, Petr
Improved variants of the FastICA algorithm.
Advances in Independent Component Analysis and Learning Machines. Londýn: Elsevier, 2015 - (Bingham, E.; Kaski, S.; Laaksonen, J.; Lampinen, J.), s. 53-74. ISBN 978-0-12-802806-3
Grant CEP: GA ČR(CZ) GA14-13713S
Institucionální podpora: RVO:67985556
Klíčová slova: independent component analysis * blind source separation * FastICA * efica * Cramer-Rao lower bound
Kód oboru RIV: BB - Aplikovaná statistika, operační výzkum
http://library.utia.cas.cz/separaty/2015/SI/tichavsky-0444036.pdf

The article presents a survey of improved variants of the famous FastICA algorithm for Independent Component Analysis. Variants of the algorithm tailored to separate mixtures of stationary non-Gaussian signals and mixtures of nonstationary (block-wise stationary) non-Gaussian signals are described. Performance analyses of the algorithms are given and compared to the respective Cramer-Rao lower bounds. The behavior of FastICA variants when additive noise is present in the signal mixture is studied through a bias analysis.
Trvalý link: http://hdl.handle.net/11104/0246783