摘要
本文给出了利用现场可编程门阵列来实现多层前向神经网络(反向传播-BP网络)的方法。首先利用了相关软件在理论上作了算法上的仿真,在此基础上构建了前向神经网络的硬件结构。主要使用了查找表的方式来实现Sigmoid激励函数,并给出了解决异或问题的硬件上的具体方案。最后给出了BP网络解决异或问题的QuartusII仿真结果,表明了方案的可行性。
A method of designing multi-layer feed-forward networks with FPGAs is provided. A type of hardware circuit for a back propagation (BP) network is constructed, which based on algorithm simulation using computational software (e.g. Matlab^TM). With the use of Look-Up-Table technique, all the activation functions in this text are Sigmoid type. A hardware neural network to resolve exclusive-OR problem is present. Finally, the simulation results of exclusive-OR problem are given, and they show that the method in this text is a feasible one.
出处
《微计算机信息》
2009年第26期215-216,共2页
Control & Automation
基金
伊犁师范学院科研计划项目(2008YB003)
关键词
现场可编程门阵列
神经网络
反向传播算法
field programmable gate arrays
neural networks
back-propagation algorithm
作者简介
郜参观(1975-),男,硕士,讲师,研究方向:数字信号处理、射频技术。通讯地址:(835000新疆伊宁市解放西路298号物电学院)