BMAC网络及其在函数学习中的应用

BMAC Neural Network and Its Application in Function Learning

论述了将n阶B样条函数运用到CMAC网络中形成BMAC网络的过程,并讨论了BMAC网络感受域函数和网络输出特性.BMAC网络与CMAC网络相比,克服了输入状态空间和输出状态空间离散化的缺点,具有连续的输入状态空间和输出状态空间.在函数学习中通过与CMAC网络的对比,体现出了BMAC网络具有逼近速度快,精度高的特性,同时也得出了学习参数及网络权值初始化对学习速度及逼近精度的影响规律.

BMAC neural network has been established by the introduction of B-splines into CMAC neural network. Ways of establishing BMAC and the specialties of BMAC output and its receptive field functions are described in details. BMAC has continuous input and output space compared with CMAC whose input and output space are dispersed. In function learning, BMAC approaches faster and is more accurate than CMAC. At the same time, rules of influence of the studying parameters and weights initialization in function learning are obtained.