The learning effect of an artificial neural network basically depends on theactivation function used for the neurons in hidden layers. In this paper, we try toanalyze the learning effect of an artificial neural network using a bell-shapedactivation fu...
The learning effect of an artificial neural network basically depends on theactivation function used for the neurons in hidden layers. In this paper, we try toanalyze the learning effect of an artificial neural network using a bell-shapedactivation function for the neurons in hidden layers. For this purpose, we present asimple neural network using a bell-shaped function, and attempt to model thecharacteristics of a dynamic system through the learning process by errorback-propagation. Through simulations, we shows the learning effect of the neuralnetwork using a bell-shaped function by comparing the case of considering asigmoid function. As a result, it is shown that training of neural networks using abell-shaped activation function is relatively fast and stable, so that such a neuralnetwork can be usefully applied for more appropriate modeling of dynamic systems.