基于FPGA的卷积神经网络浮点激励函数实现

李施豪; 应三丛

Chinese

您当前的位置：

首页 >

文章列表页 >

基于FPGA的卷积神经网络浮点激励函数实现

更新时间：2026-09-04

- 基于FPGA的卷积神经网络浮点激励函数实现
- The Implementation of Activation Function of CNN Based on FPGA Using Single Precision Floating-Point-Unit
- 微电子学与计算机 2017年34卷第10期页码：105-109
- 作者机构：
  
  四川大学计算机学院视觉合成图形图像技术国家重点学科实验室机构,四川,成都,610065
- 作者简介：
- 基金信息：
  
  国家"八六三"计划项目 2015AA016405四川省科技厅科技支撑项目 2016GZ0097
- DOI：
  中图分类号：
- 纸质出版：2017
- 稿件说明：
移动端阅览
李施豪, 应三丛. 基于FPGA的卷积神经网络浮点激励函数实现[J]. 微电子学与计算机, 2017, 34(10): 105-109.

LI Shi-hao, YING San-cong. The Implementation of Activation Function of CNN Based on FPGA Using Single Precision Floating-Point-Unit[J]. 2017, 34(10): 105-109.
李施豪, 应三丛. 基于FPGA的卷积神经网络浮点激励函数实现[J]. 微电子学与计算机, 2017, 34(10): 105-109. DOI：

LI Shi-hao, YING San-cong. The Implementation of Activation Function of CNN Based on FPGA Using Single Precision Floating-Point-Unit[J]. 2017, 34(10): 105-109. DOI：

摘要

卷积神经网络因深度学习概念的提出再一次被研究人员所重视.激励函数是卷积神经网络的一个重要组成部分

选取了sigmoid函数作为实验对象.讨论了当前几种可行的逼近方法

最终采用分段四阶多项式拟合sigmoid函数.在FPGA上使用Verilog硬件描述语言设计了并行电路

并采集了数据集进行FPGA与CPU版本caffe库进行运算效率对比.实验结果表明

此种方法误差小效率高

FPGA在深度学习领域有着广阔的应用前景.

Abstract

Convolutional Neural Network (CNN) was gotten attention because of the presentation of Deep-Learning. It has large research value for Deep-Learning in using FPGAs. The activation function is the most necessary part of CNN. In this paper

Sigmoid function was chosen as the experiment object. The approximation ways of Sigmoid Function were listed and analysed

the piecewise forth-order approximation was the best way to fit Sigmoid Function. Parallel calculation circuit was designed by Verilog HDL on FPGAs

collecting dataset and inputing them to FPGA Platform and CPU Platform. The experiment result set forth that this solution had high efficiency and low error with rosy prospect about Deep-Learning and FPGA.

关键词

Keywords

references

Djalal Eddine, Khodja Stéphane Simard. Implementation of optimized approximate sigmoid Function on FPGA circuit to use in ANN for control and monitoring[J]. CONTROL ENGINEERING AND APPLIED INFORMATICS, 2015, 17(2): 64-72. DOI: 64

Li Bingzhe, Hassan Najafi M, David J Lilja. Using stochastic computing to reduce the hardware requirements for a restricted boltzmann machine classifier[C]// The 2016 ACM/SIGDA International Symposium on Field-Programmable Gate Arrays. Monterey, California, USA, 2016: 36-41.

Vishnu P Nambiar, Mohamed Khalil-Hani, Riadh Sahnoun, et al. Hardware implementation of evolvable block-based neural networks utilizing a cost efficient sigmoid-like activation function[J]. Neurocomputing, 2014(140): 228-241. DOI: 228

LeCun Yann, Yoshua Bengio, Geoffrey Hinton. Deep learning[J]. Nature, 2015, 521(7553): 436-444. DOI: 436

EMMANUEL ADETIBA, IBIKUNLE F A, DARAMOLA S A, et al. Implementation of efficient multilayer perceptron ANN neurons on field programmable gate array chip[J]. International Journal of Engineering % Technology IJET-IJENS, 2014, 14(1): 151-159. DOI: 151

Jia Yangqing, Evan Shelhamer, Jeff Donahue, et al. Caffe: Convolutional architecture for fast feature embedding[C]// Proceedings of the 22nd ACM international conference on Multimedia. ACM, 2014.

Manish Panicker, BabuC. Efficient FPGA implementation of sigmoid and bipolar sigmoid activation functions for multilayer perceptrons[J]. IOSR Journal of Engineering, 2012, 2(6): 1352-1356. DOI: 1352

Sahil Abrol, Rita Mahajan. Implementation of single artificial neuron using various activation functions and XOR gate on FPGA chip[C]//IEEE: 2015 Second International Conference on Advances in Computing and Communication Engineering. IEEE, 2015: 118-123.

李彦冬, 郝宗波, 雷航. 卷积神经网络研究综述[J]. 计算机应用, 2016, 36(9): 2508-2515. DOI: 2508

浏览量

下载量

CSCD

文章被引用时，请邮件提醒。

提交

工具集

关联资源

一种移动卷积神经网络的FPGA实现

语音识别中卷积神经网络的FPGA实现

面向处理器分支预测单元的Spectre攻击轻量化防御方法

AlN/GaN混合缓冲层双异质结p-GaN HEMT直流特性仿真

注意力机制与改进密集卷积引导的信息隐藏模型