Blind source separation problems have recently drawn a lot of attention in
unsupervised neural learning. In the current approaches, the number of sour
ces is typically assumed to be known in advance, but this does not usually
hold in practical applications. In this paper, various neural network archi
tectures and associated adaptive learning algorithms are discussed for hand
ling the cases where the number of sources is unknown. These techniques inc
lude estimation of the number of sources, redundancy removal among the outp
uts of the networks, and extraction of the sources one at a time. Validity
and performance of the described approaches are demonstrated by extensive c
omputer simulations for natural image and magnetoencephalographic (MEG) dat
a. (C) 1999 Elsevier Science B.V. All rights reserved.