论文部分内容阅读
如何选择合适网络参数是传统 CMAC(Cerebellar Model Articulation Controller)应用中的一个难题 .采用泛化均方差 (GMSE)和学习均方差 (L MSE)来分别评价超闭球 CMAC的泛化能力与记忆精度 ,并引入权调整率的概念 ,来研究 CMAC结构参数与学习性能的关系 .研究结果表明 ,在样本分布和量化级数不变时 ,泛化均方差和学习均方差是权调整率的非增函数 .因此超闭球 CMAC在满足存储空间和计算速度的要求下尽量使得权调整率较大 .还提出了并行CMAC结构以进一步提高单个超闭球 CMAC的非线性逼近能力 .仿真结果证明了该方法的有效性
How to choose the appropriate network parameters is a difficult problem in the application of traditional CMAC (Cerebellar Model Articulation Controller) .The generalized mean square error (GMSE) and learning mean square error (L MSE) are used to evaluate the generalization ability and memory accuracy , And introduces the concept of weight adjustment rate to study the relationship between CMAC structure parameters and learning performance.The results show that the generalized mean square error and learning mean square error Function.Therefore, the super closed-ball CMAC tries to make the weight adjustment as much as possible to meet the requirements of memory space and computing speed.It also proposes a parallel CMAC structure to further improve the non-linear approximation ability of a single closed spherical CMAC.The simulation results show that The effectiveness of the method