Local Model Networks and Local Learning

Abstract

: The Local Model Networks (networks composed of locally accurate models, where the output is interpolated by smooth locally active basis functions) described in this paper provide a solid basis for practical modelling tasks. The architecture benefits from being able to incorporate Fuzzy, Neural Network and conventional System Identification methodology and experience. The advantages of the architecture are described, and the tradeoff between Local and Global Learning is investigated. The Local Learning method is computationally less expensive and was found to lead to smoother and more interpretable solutions than global learning. The results are illustrated with a robot actuator modelling problem. 1. Introduction Modelling nonlinear dynamic systems from observed data and a priori engineering knowledge is a major area of science and engineering. In recent years a great deal of work has appeared in new areas like Fuzzy Modelling and Neural Networks to complement the previous work in st...