Neural Networks for Blind Separation with Unknown Number of Sources

摘要

Blind source separation problems have recently drawn a lot of attention in211u001eunsupervised neural learning. In this paper, various neural network architectures 211u001eand associated adaptive learning algorithms are discussed for handling the cases 211u001ewhere the number of sources is unkown. These techniques include estimation of the 211u001enumber of sources, redundancy removal among the outputs of the networks, and 211u001eextraction of the sources one at a time. Validity and performance of the 211u001edescribed approaches are demonstrated by extensive computer simulations for 211u001enatural image and magnetoencephalographic (MEG) data.