Dynamics of feedback neural nets with unsupervised learning

The dynamics of feedback artificial neural networks (ANNs) with unsupervised learning is examined in order to describe their inherent genetic characteristics. A simple prototype continuous-time feedback ANN with unsupervised learning is considered and its inherent characteristics are described. Based on the choice of the genetic parameters, the prototype can learn only one class, only two classes, or only three classes. To which class an initial external datum would converge, however, would also depend on the initial value of the weight. Some of the conclusions are extended in order to describe qualitatively the characteristics of an arbitrarily large feedback ANN with unsupervised learning