Learning performance of a neurocomputer for nonlinear dynamical system identification

This paper investigates the learning performance of a RICOH neurocomputer RN-2000 for the identification problem of input and output map of a discrete nonlinear dynamical system. The results obtained show capability of on-chip learning, which is essential for many neural applications such as machine learning and control where real-time adaptation is required. In this paper, the method to use a neurocomputer is briefly presented for a nonlinear identification problem. The main significance of this research is to obtain a further guideline for designing a primitive artificial brain for robotics.