Title :
Bayesian Ying-Yang learning based ICA models
Author_Institution :
Dept. of Comput. Sci. & Eng., Chinese Univ. of Hong Kong, Shatin, Hong Kong
Abstract :
It has been shown that a particular case of the Bayesian Ying-Yang learning system and theory will reduce into a very general ICA framework. It not only includes the existing information-theoretic ICA approaches as particular examples, but also improve their performances extend them to handle the cases that sensors are affected by noises and outliers and cases that the number of sensors is larger than the number of sources, and also be able to detect the correct number of sources. Algorithms are developed for implementing this ICA framework both in its general form and in its simplified versions for two important special cases, supported by some theoretical results and experimental demonstration
Keywords :
Bayes methods; information theory; learning (artificial intelligence); neural nets; signal resolution; Bayesian Ying-Yang learning; ICA models; independent component analysis models; information-theoretic approaches; noises; outliers; Bayesian methods; Computer science; Density functional theory; Entropy; Independent component analysis; Learning systems; Mutual information; Neural networks; Statistical learning; Unsupervised learning;
Conference_Titel :
Neural Networks for Signal Processing [1997] VII. Proceedings of the 1997 IEEE Workshop
Conference_Location :
Amelia Island, FL
Print_ISBN :
0-7803-4256-9
DOI :
10.1109/NNSP.1997.622429
