Kurtosis extrema and identification of independent components: a neural network approach

We propose a nonlinear self-organising network which solely employs computationally simple hebbian and antihebbian learning in approximating a linear independent component analysis (ICA). Current neural architectures and algorithms which perform parallel ICA are either restricted to positively kurtotic data distributions [1] or data which exhibits one sign of kurtosis [2, 3, 12]. We show that the proposed network is capable of separating mixtures of speech, noise and signals with both platykurtic (positive kurtosis) and leptokurtic (negative kurtosis) distributions in a blind manner. A simulation is reported which successfully separates a mixture of twenty sources of music, speech, noise and fundamental frequencies.