用多个改进低维CMAC神经网络学习多维函数的研究

Research on Learning a Multidimensional Function Using Many Modified Low Dimensional CMAC Neural Networks

提出了一种多维非线性函数的多神经网络学习方法，即用变量代换的方法把一个多维非线性函数分解为若干低维函数，用多个改进的低维小脑模型神经网络分别映射这些低维函数，提高了收敛性，减少了存储空间，大大提高了学习精度，且易于实现。给出了大量学习非线性函数的仿真实验, 其结果表明，采用这种方法的学习精度比用一个CMAC的学习精度提高10倍以上。

This paper describes a method of learning a multidimensional nonlinear function using many modified low dimensional cerebellar model articulation controller(CMAC), By this method, a multidimensional nonlinear function can be decomposed to many low -dimensional functions by variate substitute , then, low -dimensional functions are mapped using many modified low dimensional CMAC, thus, the convergencing are improved and storage space are lessened and the accuracy of learning are greatly improved and it can be easily realized . The computer experiments demonstrate that the learning accuracy are improved tenfold more than the accuracy using a CMAC.