engleski

Dynamic neural network with adaptive Gauss neuron activation function

An attempt has been made to establish a nonlinear dynamic discrete-time neuron model, the so called Dynamic Elementary Processor (DEP). This dynamic neuron disposes of local memory, in that it has dynamic states. To accelerate the convergence of proposed extended dynamic error-back propagation learning algorithm, the adaptive neuron activation function and momentum method are applied. Instead of most popular bipolar and unipolar Sigmoid neuron activation functions, the Gauss activation function with adaptive parameters is proposed. Based on the DEP neuron with adaptive activation function in hidden layer, a Dynamic Multi Layer Neural Network is proposed and tested in prediction of a Glass-Mackey time series. The learning results are presented in terms that are insensitive to the learning data range and allow easy comparison with other learning algorithms, independent of machine architecture or simulator implementation.

dynamic neural network, adaptive neuron activation function, momentum, prediction, Glass-Mackey time series

nije evidentirano