engleski

Modified Dynamic Neuron Model

In this paper a modification of the nonlinear dynamic discrete-time neuron model, the so-called Dynamic Elementary Processor (DEP), is proposed. DEP disposes of local memory, in that it has dynamic states. Instead of the most popular unipolar and bipolar Sigmoidal neuron activation functions, the Gauss activation function with adaptive parameters is applied. Based on the DEP neurons in hidden layer a modified dynamic neural network (MDNN) without any Bias neurons is proposed. For such neural network, the Error Back- Propagation and RPROP learning strategies are compared in solving of two benchmarks, the Glass-Mackey time series prediction and XOR classification problem.

dynamic neural network; adaptive activation function; bias neuron; error back-propagation; RPROP

nije evidentirano