Comparison of complex- and real-valued feedforward neural networks in their generalization ability

Abstract

We compare the generalization characteristics of complex-valued and real-valued feedforward neural networks when they deal with wave-related signals. We assume a task of function approximation. Experiments demonstrate that complex-valued neural networks show smaller generalization error than real-valued ones in particular when the signals have high degree of wave nature. © 2011 Springer-Verlag.