A Hybrid Neuro-Fuzzy Element: A New Structural Node for Evolving Neuro-Fuzzy Systems

摘要

A modification of the structure for a neurofuzzy unit was offered which is generally a hybrid system that combines nonlinear synapses and an activation function to form the hybrid system's output value. The introduced neurofuzzy element is specifically an extension of the common neo-fuzzy neuron which is upgraded at the expense of application of an additional (contracting) activation function. A particular robust learning procedure is also considered for this case that makes it possible to reduce errors while processing data containing abnormal observations.