摘要
提出了一种适合集群计算机上实现的基于模型的并行神经网络训练算法,分析了其代价函数和加速比,在作者研制的集群计算机上BSP进行了实现和测试,结果表明:当问题规模比较大时,此并行算法能获得好的加速比。
In this paper,we present an artificial neural network algorithm Based on BSP (bulk synchronous parallel) which can be implemented on PC cluster. We investigate the cost function and speedup of this algorithm. An implementation has been tested on PC cluster developed by using the Oxford BSPlib. The results show that the BSP algorithm is of great value to a large scale ANN problem in practice.
出处
《计算机工程》
CAS
CSCD
北大核心
2002年第1期59-61,共3页
Computer Engineering
基金
国家自然科学基金项目(69775012)