In the article it is given the comparative analysis of recovering noise-distorted multidimensional functions by CMAC, RBF and MP neural networks of different architectures. The problems of effective selecting the parameters of these neural networks and CMAC neural networks basis functions are considered. It is shown that CMAC-based neural networks allow achieving the lowest training times; and using RBF and MP networks provide the estimated recovery accuracy requiring less memory but considerably higher computational costs.
"O neirosetevom podkhode k vosstanovlenyiu mnohomernыkh funktsyi pry nalychyy pomekh yzmerenyi" ,
Information Processing Systems,