摘要
精度要求是神经网络硬件实现的一个关键问题。本文提出了一种有限精度的BP高速算法，通过实例验证，该算法适于低精度，且收敛速度快。
The precision required for neural network algorithm is an important question for hardware architects. This paper gives an excellent high speed BP algorithm for feed-forward muitilayerneural network using limited precision calculation. The experimental results show that the algorithmimproves convergence speed with reduced precision bits.
出处
《国防科技大学学报》
EI
CAS
CSCD
1998年第1期88-91,共4页
Journal of National University of Defense Technology
关键词
BP学习算法
代价函数
有限精度
硬件
BP神经网络
back-propagation, feed-forward multilayer neural network, cost function, activationfunction, limited precision calculation, neural network hardware
