Hash-coding in CMAC neural networks

Abstract

Hash-coding is used in CMAC neural networks to reduce the required memory, thereby making the CMAC practical to implement. In this paper the original motivation and rationale for using hash-coding in CMAC are questioned and it is shown that, contrary to the traditional believe, that hash-coding is unable to enhance CMAC's approximation ability. A comparison between the CMAC performance obtained with two popular hash-coding methods is given.