A novel CMAC-type neural network based on Newton´s forward interpolation

This paper proposes a novel CMAC neural network based on Newton´s forward interpolation: NFI-CMAC, which is capable of implementing error-free approximations to multi-variable polynomial functions of arbitrary order, including Newton´s forward interpolation polynomials, conceptual receptive field, content-addressing mechanism, and training algorithm. The advantages it offers over conventional CMAC neural network are: high-precision of learning, much smaller memory requirement without the data-collision problem, much less computational effort for training and faster convergence rates than that attainable with multi-layer BP neural networks. A set of numerical simulations have been conducted, and simulation results have shown that the novel neural network is feasible and efficient in function approximation. The novel neural network has great potential in the application areas of signal processing, pattern recognition, process modeling and implementation of high-precision real-time intelligent controller.