一种神经网络硬件实现的可重构设计

Reconfigurable design of neural networks hardware implementation

以BP网络为例,提出了一种可重构神经网络硬件实现方法。通过可重构体系结构、可重构部件的设计,可以灵活地实现不同规模、传递函数及学习方法的神经网络,从而搭建起神经网络快速硬件实现的平台。经过对一个模式识别问题的实现和测试,证明了这种设计方法的可行性。

The BP networks was taken as an example and a reconfigurable method of neural networks（NN） hardware implementation was proposed. Based on this reconfigurable architecture and components, NN with different scales, transfer functions or learning algorithms could be implemented flexibly and fast. The implementation and test of a pattern recognition problem prove the feasibility of this method.