模块神经网络及其性能  

MODULAR NEURAL NETWORK AND ITS PERFORMANCE

作　　者：王宝翰[1] 陈劲 

机构地区：[1]中国科学院生物物理研究所

出　　处：《生物物理学报》1994年第4期671-680,共10页Acta Biophysica Sinica

摘　　要：提出一个不同于实现ｙ＝ｆ（ｘ）的ＢＰ网络的神经网络模型，给出了网络的结构及并行动力学方程，证明了其动力学的稳定性。通过学习算法的建立，证明网络能精确实现输入矢量对（ｘ，ｙ）映入成相联系的输出矢量ｚ，最重要的是网络能同时存诸依时变化的时序模式与静态模式。此外并给出动力学学习算法，证明此学习算法的收敛性，计算机仿真证实理论结果，最后讨论了某些可能的应用。A multi-modular neural network model, which is different from BP network for mapping y=f(x),is proposed. The architecture and parallel dynamics equations of the network are given,and the stability of dynamics is also proven. Through establishing learning algorithm. We have proven that the network can perform mapping or input vector pair (x,y) to associative output vector z. It is the most inportant that both the pattern series varing with the and statics patterns can be stored sitmutaneously, thus the economic principle of memory is proposed. In addition, a dynamics learning algorithm is given and have proven its convergence, computer simulation confirms fully theoretical results. Finally, Some possible applications are discussed.

关 键 词：模块 神经网络 结构 人工智能 

分 类 号：Q811.211[生物学—生物工程] TP18[自动化与计算机技术—控制理论与控制工程]