An overview of the CMAC neural network

Abstract

The authors describe the cerebellar model arithmetic computer (CMAC) neural network, which is an alternative to backpropagated multilayer networks. CMAC has properties of generalization, rapid algorithmic computation based on least-mean-square (LMS) training, functional representation, output superposition, and practical hardware realization, all of which are discussed. Data concerning CMAC capacity and generalization are shown. Brief descriptions of applications in pattern recognition, robot control, and signal processing are given.<>