Artificial neural networks with adaptive polynomial activation function

The aim of this work is to study an extended multilayer perceptron made of neurons with an adaptive polynomial activation function. The adaptive polynomial neural network (APNN) gives a reduction in terms of dimensions and omputational complexity both in learning and in forward phase compared with traditional MLPs with a sigmoidal activation function. Many experiments have been extensively carried out both on pattern recognition and data processing problems. The relationship of the APNNs with the Volterra expansion ...