Local Learning for S_CMAC_GBF

Abstract

This paper proposed a method of the local learning for S_CMAC_GBF. CMAC_GBF performs better than CMAC in the learning convergence speed and accuracy. The advantages of S_CMAC_GBF are simpler addressing structure, lower consuming memory space and easier hardware application, and it will not degrade the accuracy of the original input. For achieving better learning efficiency and speeding the hardware output, this paper proposed the local learning to perform more efficient learning. In local learning, every area only use better relative weights rather than use all weights, so the computation can be speed up in learning. Because in the hardware output, it only needs to calculate better relative weights, therefore, the learning speed is greatly increased. Although local learning lowers down the output accuracy of S_CMAC_GBF, the advantage of using S_CMAC_GBF is not affected, especially in the hardware application; the accuracy is almost the same. In the paper, a FPGA chip is employed to demonstrate the better hardware efficiency.