The research of approximative possibilities of radial-basis networks with different activation functions

The article describes an automated system developed by the authors to approximate the probability density function of random processes. This system uses radial-basis network (RBF), which uses kernel functions as the activation of neurons in addition to the Gaussian functions. The article analyzes the results of approximation produced by networks with different activation functions of neurons and concludes an advisability of replacing the Gaussian functions by kernel functions.