摘要
小脑模型清晰度控制器(CMAC)是一种局部学习前馈网络,结构简单,收敛速度快,易于实现。从其每个神经元来看,各神经元之间是一种线性关系,但从总体结构来看,网络是一种非线性映射关系。而且模型从输入开始就存在一种泛化能力。网络的学习和泛化能力一直是研究热点,因此,该文将对CMAC网络的泛化能力、学习能力以及一些改善途径进行多方面的综合性的讨论。文章最后还将给出一种改善CMAC泛化能力的训练策略,它不仅避免了学习干扰问题加快了学习速度而且可以通过提高训练循环次数增加训练样本量。通过MATLAB仿真发现这种训练策略可以改善CMAC网络的泛化能力。该方法简单有效是可行的。
CMAC ( Cerebellar Model Articulation Controller) is a kind of local learning feed - forward neural network with simple architecture, quick learning convergence and effective implementation. Although it is linear between different nerve cells, the mapping of the whole network is nonlinear. Furthermore a kind of generalization capability is produced along with the input. The learning and generalization capability always attracts much attention of researchers. In this paper, the learning, generalization capability and some improvements are introduced synthetically. In the end, a training strategy is presented which can improve the generalization capability. This strategy not only removes the learning interference and quickens learning speed but also increases sample quantity by advancing training cycle times. Improvement of the generalization capability is proved by the MATLAB simulation. The method is simple, effective and feasible.
出处
《计算机仿真》
CSCD
2005年第6期5-7,31,共4页
Computer Simulation