Research on neural network pruning and approximate computing technology based on neuron fault tolerance analysis

WANG Xu; WANG Jing; ZHANG Wei-gong

Chinese

您当前的位置：

首页 >

文章列表页 >

Research on neural network pruning and approximate computing technology based on neuron fault tolerance analysis

更新时间：2026-09-04

- Research on neural network pruning and approximate computing technology based on neuron fault tolerance analysis
- Vol. 36, Issue 11, Pages: 70-75(2019)
- 作者机构：
  
  1. 首都师范大学机构信息工程学院,北京,100048
  2. 中国科学院计算技术研究所体系结构重点实验室,北京,100086
  3. 电子系统可靠性北京市重点实验室,北京,100048
- 作者简介：
- 基金信息：
- DOI：
  CLC：
- Published：2019
- 稿件说明：
移动端阅览
王旭, 王晶, 张伟功. 基于神经元容错度分析的神经网络裁剪与近似计算技术研究[J]. 微电子学与计算机, 2019, 36(11): 70-75, 83.

WANG Xu, WANG Jing, ZHANG Wei-gong. Research on neural network pruning and approximate computing technology based on neuron fault tolerance analysis[J]. 2019, 36(11): 70-75.
王旭, 王晶, 张伟功. 基于神经元容错度分析的神经网络裁剪与近似计算技术研究[J]. 微电子学与计算机, 2019, 36(11): 70-75, 83. DOI：

WANG Xu, WANG Jing, ZHANG Wei-gong. Research on neural network pruning and approximate computing technology based on neuron fault tolerance analysis[J]. 2019, 36(11): 70-75. DOI：

摘要

本文将神经元裁剪和近似计算技术相结合

首先提出基于统计排序的神经元容错能力量化方法.然后

为了识别神经元的裁剪度

根据神经元的容错能力提出神经元重要程度排序算法.其次

引入轻量级的重训练

提出循环裁剪法

以探寻最优裁剪率.最后

根据神经元的容错能力

在神经网络运行过程中使用近似计算技术进一步降低功耗开销.本文通过两个实验

证明了该技术的有效性

其中以MNIST为例

在精度损失小于5%的情况下

压缩率达到50%

节能1.35倍.

Abstract

This paper proposes to use neuron node pruning and approximate computing simultaneously. First

we propose a method to quantify the fault tolerance capability of neurons based on statistics. Then

to identify whether the neuron can be pruned

an importance ranking algorithm is proposed based on the fault tolerance capability. Next

introducing retrain and cyclic pruning to find the optimal pruning rate. Finally

approximate computing technique is used to further reduce power consumption during neuron network execution. The effectiveness of above technique is proved by two experiments. In the case of MNIST dataset

the compression rate is 50% and the power saving is 1.35×when the output accuracy loss is less than 5%.

关键词

Keywords

references

ZHANG Q, WANG T, TIAN Y, et al. ApproxANN: an approximate computing framework for artificial neural network[C]// Design, Automation % Test in Europe Conference % Exhibition. Grenoble, France, EDA Consortium, 2015.

邢静宇, 张立臣.动态电压调整多处理器实时系统任务调度[J].微电子学与计算机, 2006, 23(2):55-57. DOI: 55

YU J, LUKEFAHR A, PALFRAMAN D, et al. Scalpel: customizing DNN pruning to the underlying hardware parallelism[J]. AcmSigarch Computer Architecture News, 2017, 45(2):548-560. DOI: 548

朱新召, 胡哲琨, 周莉, 等.基于多核处理器的多层感知神经网络设计和实现[J].微电子学与计算机, 2014, 31(11):27-31. DOI: 27

SARWAR S S, VENKATARAMANI S, RAGHUNATHAN A, et al. Multiplier-less artificial neurons exploiting error resiliency for energy-efficient neural computing[C]// 2016 Design, Automation % Test in Europe Conference % Exhibition (DATE). Dresden, Germany, IEEE, 2016.

COURBARIAUX M, BENGIO Y, DAVID J P. BinaryConnect: training deep neural networks with binary weights during propagations[C]// International Conference on Neural Information Processing Systems. Istanbul, Turkey, 2015.

GUERREIRO J, ILIC A, ROMA N, et al. GPGPU power modeling for multi-domain voltage-frequency scaling[C]// IEEE International Symposium on High Performance Computer Architecture. Vienna, Austria, 2018.

YANG T J, CHEN Y H, SZE V. Designing energy-efficient convolutional neural networks using energy-aware pruning[C]// Computer Vision % Pattern Recognition. Honolulu, HI, USA, IEEE, 2017.

CARLSON T E, HEIRMAN W, EECKHOUT L. Sniper: exploring the level of abstraction for scalable and accurate parallel multi-core simulation[C]// High Performance Computing, Networking, Storage % Analysis. Washington, USA, IEEE, 2011.

HAN S, MAO H, DALLY W J. Deep compression: compressing deep neural networks with pruning, trained quantization and huffman coding[J]. Fiber, 2015, 56(4):3-7. DOI: 3

BABAEIZADEH M, SMARAGDIS P, CAMPBELL R H. NoiseOut: a simple way to prune neural networks[J]. arXiv.org > cs > arXiv:1611.06211v1, 2016.

Views

下载量

CSCD

Alert me when the article has been cited

提交

Tools

Publicity Resources

A feature-oriented context understanding and slice analysis network for tuberculosis detection

Steganography model guided by attention mechanism and improved dense convolution

A review of large model development: technological evolution from BERT to Deepseek

Research on assembly action recognition based on MSS-YOLO-MPHands fusion architecture

Implementation and verification of a dual-chain power gating strategy for low power design

Related Author

No data

Related Institution

No data

AI问答

Postal code：100079
Tel：（010）53879206 Email：tmw@bjxintong.com.cn
Technical support is provided by Beijing Founder electronics co., LTD 京ICP备09082226号-64 京公网安备11010602201714号
It is recommended to read the content of this site in Chrome&IE9+. Please switch to extreme mode in browser 360.
Cookies We use cookies to help provide and enhance our service and tailor content. By continuing, you agree to the use of cookies.

⁰