The dendritic neuron model is a universal approximator

Abstract

According to the biological composition of neurons, the DNM was invented based on the dendrites of neurons, which can break through the functions that traditional simple neuron model (McCulloch and Pitts, 1943) cannot achieve. While by using traditional single-layer model, the approximation to any continuous function cannot be achieved by a single neuron. In this paper, an accurate mathematical method is used to demonstrate that the DNM can approximate any continuous function which could be achieved with only one neuron.