Multi{Layer Neural Networks with a Local Adaptive Learning Rule for Blind Separation of Source Signals

In this contribution a class of simple local unsupervised learning algorithms is proposed for multi{ layer neural network performing source signal separation from linear mixture of them (the blind separation problem). The main motivation for using a multi{layer network instead of a single layer one for the blind separation problem is to improve the performance and robustness of separation while applying local learning rules. These rules are biologically justi ed opposite to existing more complex global learning rules. The proposed algorithms allow the separation of badly scaled signals and in case of ill{conditioned problems (if very similar mixtures of sources are available only). The application of developed methods for image enhancement is demonstrated.