论文部分内容阅读
本文对成批训练BP算法原有串行程序进行了数据相关性分析,提出了三种并行划分方法,在此基础上实现了以EP-860并行计算机系统为运行环境的并行计算程序.文章对三种划分方法从通信复杂性、处理机的空闲等待率等几个方面作了比较,最后给出了不同划分方法的使用场合.
This paper analyzes the data correlation of the original serial programs of batch training BP algorithm and puts forward three parallel methods of division, on this basis, realizes the parallel computing program based on the EP-860 parallel computer system as the operating environment. The three methods are compared from the aspects of communication complexity and idle waiting rate of the processor. Finally, the application situations of different methods of division are given.