Unsupervised neural network to nonlinear blind separation

Summary form only given. Nonlinear blind separation has received much research attention recently due to the emergence of simple, powerful algorithms that show promise in practical applications. In this paper, we consider a nonlinear mixture model. We use unsupervised neural network self-organizing maps (SOM), by applying expectation-maximization (EM) as a learning algorithm for finding the sources. The EM algorithm yields topology preserving maps of data based on probabilistic mixture models. Our approach has the benefits of both EM and SOM algorithms, without constraints on source signals.