基于FPGA的Gaussian传输函数实现

FPGA-based Implementation of Gaussian Transfer Function

Gaussian函数作为RBF神经网络常用的激活函数,在FPGA上实现Gaussian函数,有利于提高神经网络的运算速度。本文简要介绍了Gaussian函数模型,并基于FPGA平台,分析了几种实现复杂函数的方案,最终采用分段拟合的方法对Gaussian函数进行逼近。首先用MATLAB对Gaussian函数进行初步的拟合,在FPGA平台用Verilog HDL语言实现Gaussian函数。仿真结果表明,误差可以控制在较小的数量级,满足神经网络运算的精度。

As a commonly used activation function of RBF neural network, Gaussian function realizes Gaussian function on FPGA, which is beneficial to improve the operation speed of neural network. This paper briefly introduces the Gaussian function model, and based on the FPGA platform, analyzes several schemes for implementing complex functions. Finally, the Gaussian function is approximated by the piecewise fitting method.Firstly, the Gaussian function is firstly fitted with MATLAB, and the Gaussian function is implemented in the FPGA platform with Verilog HDL language. The simulation results show that the error can be controlled to a small order of magnitude, which satisfies the accuracy of neural network operations.