Nonsymmetric PDF approximation by artificial neurons: application to statistical characterization of reinforced composites

Abstract

We present a generalized adaptive activation function neuron structure which learns through an information-theoretic-based principle, which is able to blindly estimate the probability density function of incoming input. We illustrate the behavior of the learning theory by the help of numerical experiments performed on real-world data with particular emphasis to statistical characterization of polypropylene composites reinforced with vegetal fibers.