Neural networks

The possibility of predicting the regular and chaotic dynamics of a discrete map by using artificial neural network is studied. The method of error back­propagation is used for calculation the coefficients of the multilayer network. The predicting properties of the neural network are explored in a wide region of the system parameter for both regular and chaotic behaviors. The dependance of the prediction accuracy from the degree of chaos and from the number of layers of the network is studied.

The conceptual model and computer simulations results of path integration in free­scalable nonlinear oscillator neural networks with even cyclic inhibition (ECI­networks) are discussed in this paper. To estimate the phase shifting under input impact the ECI­networks contain two subsystems namely reference and information ones. The population of reference (nonencoding) oscillatory units has significant role in generation and stabilization of numerous time scales despite it don’t assist directly in the phase pattern encoding of input signals.