Information-theoretic learning for FAN network applied to eterokurtic component analysis

Abstract

The paper presents a novel approach for performing independent component analysis of mixed plati-kurtic and lepto-kurtic source signals, which is referred to as the ‘eterokurtic’ blind source separation problem. The approach employs a neural network formed by adaptive activation function neurons, which provide the statistics required for learning by the extended INFOMAX theory. Through computer simulations conducted on both synthetic and real-world data, the proposed approach is assessed and its effectiveness is illustrated.