Nessy - An Evolutionary Learning Neural Network

A neural network is presented, which uses evolutionary learning. The neural network architecture is based on the multilayer backpropagation (MBPN) architecture, the evolutionary learning is based on genetic algorithms (GA). Both, GA and NN, are neatly integrated into the Nessy architecture. Every neuron of Nessy has a chromosome of the GA as its state. One cycle of the neural network algorithm represents a generation of the GA. The algorithm of Nessy has five parts: selection, error computation, weights modification, transduction and mutation. The Nessy algorithm was examined by using a complex test function, and some of its basic learning properties are derived. It shows, that the Nessy learning algorithm performs faster and more reliable learning than conventional GA. This is a result from the use of both, the weights as additional "memory", and the transduction operator as replacement for the cross-over operator.