论文部分内容阅读
神经网络实现技术是神经网络研究的一个极重要的领域。本文首先分析了神经网络模拟对并行计算机系统的要求,认为影响神经网络计算机速度和容量提高的主要因素是单个处理单元的速度、单个处理单元的局部存储器的容量以及互连网络的通信带宽。要提高模拟神经计算机的速度和容量,就要有相应的并行结构来支持。在定量的需求分析的基础上,本文还提出了一种模拟神经计算机的并行结构。
Neural network technology is a very important area of neural network research. In this paper, the requirements of parallel computer system for neural network simulation are analyzed. The main factors affecting the speed and capacity of neural network computer are the speed of single processing unit, the capacity of local memory of single processing unit and the communication bandwidth of interconnection network. To increase the speed and capacity of a simulated nerve computer, there should be a corresponding parallel structure to support it. On the basis of quantitative demand analysis, this paper also presents a parallel structure simulating neural computer.