0041066 20240103182727.220060913235959.9 13 s. cav_un_epca*02532421045-9227175 (2006)1265-1277 Independent component analysis blind source separation blind deconvolution Cramer-Rao lower bound algorithm FastICA cav_un_auth*0108100 Koldovský Zbyněk UTIA-B Department of Stochastic Informatics Ústav teorie informace a automatizace AV ČR, v. v. i. cav_un_auth*0101212 Tichavský Petr UTIA-B Department of Stochastic Informatics Ústav teorie informace a automatizace AV ČR, v. v. i. cav_un_auth*0213223 Oja E. FI 12B 1M0572 GA MŠk cav_un_auth*0001814 CEZ:AV0Z10750506 FastICA is one of the most popular algorithms for Independent Component Analysis, demixing a set of statistically independent sources that have been mixed linearly. A key question is how accurate the method is for finite data samples. We propose an improved version of the FastICA algorithm which is asymptotically efficient, i.e., its accuracy given by the residual error variance attains the Cram'er-Rao lower bound. The error is thus as small as possible. This result is rigorously proven under the assumption that the probability distribution of the independent signal components belongs to the class of generalized Gaussian distributions with parameter $/alpha$, denoted GG$(/alpha)$ for $/alpha >2$. We name the algorithm EFICA. Computational complexity of a Matlab$^TM$ implementation of the algorithm is shown to be only slightly (about three times) higher than that of the standard symmetric FastICA. Algoritmus FastICA je jednim z popularnich algoritmu ktere slouzi ke slepe separaci puvodne nezavislych signalu, ktere byly linearne smichane dohromady.V clanku je navrzena vylepsena varianta tohoto algoritmu, ktera je statisticky eficientni, tj. jeji presnost merena pomoci rezidualni variance chyby dosahuje Rao-Cramerovy meze. Tento vysledek j eodvozen za predpokladu, ze pravdepodobnostni distribuce puvodnich signalu patri do rodiny zobecnenych Gaussovych distribuci. Vypocetni narocnost nove procedury jen mirne (asi trikrat) prevysuje slozitost symetricke varianty algoritmu FastICA. Vlastnosti algoritmu jsou porovnavany v simulacich s jinymi algoritmy, a to nejen na tride zobecnenych Gaussovskych distribucich ale take na bi-modalnich distribucich a na separaci linearne smichanych recovych signalu. 2007 BB http://hdl.handle.net/11104/0134652 2006 cav_un_epca*0253242 IEEE Transactions on Neural Networks 1045-9227 Roč. 17 č. 5 2006 1265 1277