论文部分内容阅读
1.引言点格神经网络为仿真离散空间的时空动力学系统提供了一个可行性框架,特别是其联接方式比Hopfield型的网络简单得多,因此从更为重要的实用观点来看,它能提供模拟阵列计算机结构的VLSI实现,从而可构造一种通用的模逻(Analogic)计算机,借以解决各种科学和工程问题。用于求解偏微分方程的点格神经网络所描述的是一个空间离散时间动力学系统,其行为非常类似于空间连续系统。虽然空间离散不可避免地会引入一些误差,但在其它方面两个系统是定性相同的:它们均为连续时间动力学系统且状态变量、相互作用
1. INTRODUCTION The point-lattice neural network provides a feasible framework for simulating the space-time dynamics of discrete space. Especially, the connection method is much simpler than the Hopfield-type network. Therefore, from a more important practical point of view, it can Provides a VLSI implementation that mimics the structure of an array computer so that a versatile analogic computer can be constructed to solve a variety of scientific and engineering problems. The point-lattice neural network used to solve partial differential equations describes a system of discrete-time dynamics that behaves much like a spatially continuous system. Although spatial discretization inevitably introduces some errors, the other two systems are qualitatively the same: they are both continuous-time dynamical systems and state variables, interactions