通过运行时层冻结、模型量化和提前停止实现高能效神经网络训练

IF 4.1 2区 计算机科学 Q1 COMPUTER SCIENCE, HARDWARE & ARCHITECTURE
Álvaro Domingo Reguero , Silverio Martínez-Fernández , Roberto Verdecchia
{"title":"通过运行时层冻结、模型量化和提前停止实现高能效神经网络训练","authors":"Álvaro Domingo Reguero ,&nbsp;Silverio Martínez-Fernández ,&nbsp;Roberto Verdecchia","doi":"10.1016/j.csi.2024.103906","DOIUrl":null,"url":null,"abstract":"<div><h3>Background:</h3><p>In the last years, neural networks have been massively adopted by industry and research in a wide variety of contexts. Neural network milestones are generally reached by scaling up computation, completely disregarding the carbon footprint required for the associated computations. This trend has become unsustainable given the ever-growing use of deep learning, and could cause irreversible damage to the environment of our planet if it is not addressed soon.</p></div><div><h3>Objective:</h3><p>In this study, we aim to analyze not only the effects of different energy saving methods for neural networks but also the effects of the moment of intervention, and what makes certain moments optimal.</p></div><div><h3>Method:</h3><p>We developed a novel dataset by training convolutional neural networks in 12 different computer vision datasets and applying runtime decisions regarding layer freezing, model quantization and early stopping at different epochs in each run. We then fit an auto-regressive prediction model on the data collected capable to predict the accuracy and energy consumption achieved on future epochs for different methods. The predictions on accuracy and energy are used to estimate the optimal training path.</p></div><div><h3>Results:</h3><p>Following the predictions of the model can save 56.5% of energy consumed while also increasing validation accuracy by 2.38% by avoiding overfitting.The prediction model developed can predict the validation accuracy with a 8.4% of error, the energy consumed with a 14.3% of error and the trade-off between both with a 8.9% of error.</p></div><div><h3>Conclusions:</h3><p>This prediction model could potentially be used by the training algorithm to decide which methods apply to the model and at what moment in order to maximize the accuracy-energy trade-off.</p></div>","PeriodicalId":50635,"journal":{"name":"Computer Standards & Interfaces","volume":"92 ","pages":"Article 103906"},"PeriodicalIF":4.1000,"publicationDate":"2024-08-09","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"https://www.sciencedirect.com/science/article/pii/S0920548924000758/pdfft?md5=9fe0b023bddfc875c825b0c53e63af06&pid=1-s2.0-S0920548924000758-main.pdf","citationCount":"0","resultStr":"{\"title\":\"Energy-efficient neural network training through runtime layer freezing, model quantization, and early stopping\",\"authors\":\"Álvaro Domingo Reguero ,&nbsp;Silverio Martínez-Fernández ,&nbsp;Roberto Verdecchia\",\"doi\":\"10.1016/j.csi.2024.103906\",\"DOIUrl\":null,\"url\":null,\"abstract\":\"<div><h3>Background:</h3><p>In the last years, neural networks have been massively adopted by industry and research in a wide variety of contexts. Neural network milestones are generally reached by scaling up computation, completely disregarding the carbon footprint required for the associated computations. This trend has become unsustainable given the ever-growing use of deep learning, and could cause irreversible damage to the environment of our planet if it is not addressed soon.</p></div><div><h3>Objective:</h3><p>In this study, we aim to analyze not only the effects of different energy saving methods for neural networks but also the effects of the moment of intervention, and what makes certain moments optimal.</p></div><div><h3>Method:</h3><p>We developed a novel dataset by training convolutional neural networks in 12 different computer vision datasets and applying runtime decisions regarding layer freezing, model quantization and early stopping at different epochs in each run. We then fit an auto-regressive prediction model on the data collected capable to predict the accuracy and energy consumption achieved on future epochs for different methods. The predictions on accuracy and energy are used to estimate the optimal training path.</p></div><div><h3>Results:</h3><p>Following the predictions of the model can save 56.5% of energy consumed while also increasing validation accuracy by 2.38% by avoiding overfitting.The prediction model developed can predict the validation accuracy with a 8.4% of error, the energy consumed with a 14.3% of error and the trade-off between both with a 8.9% of error.</p></div><div><h3>Conclusions:</h3><p>This prediction model could potentially be used by the training algorithm to decide which methods apply to the model and at what moment in order to maximize the accuracy-energy trade-off.</p></div>\",\"PeriodicalId\":50635,\"journal\":{\"name\":\"Computer Standards & Interfaces\",\"volume\":\"92 \",\"pages\":\"Article 103906\"},\"PeriodicalIF\":4.1000,\"publicationDate\":\"2024-08-09\",\"publicationTypes\":\"Journal Article\",\"fieldsOfStudy\":null,\"isOpenAccess\":false,\"openAccessPdf\":\"https://www.sciencedirect.com/science/article/pii/S0920548924000758/pdfft?md5=9fe0b023bddfc875c825b0c53e63af06&pid=1-s2.0-S0920548924000758-main.pdf\",\"citationCount\":\"0\",\"resultStr\":null,\"platform\":\"Semanticscholar\",\"paperid\":null,\"PeriodicalName\":\"Computer Standards & Interfaces\",\"FirstCategoryId\":\"94\",\"ListUrlMain\":\"https://www.sciencedirect.com/science/article/pii/S0920548924000758\",\"RegionNum\":2,\"RegionCategory\":\"计算机科学\",\"ArticlePicture\":[],\"TitleCN\":null,\"AbstractTextCN\":null,\"PMCID\":null,\"EPubDate\":\"\",\"PubModel\":\"\",\"JCR\":\"Q1\",\"JCRName\":\"COMPUTER SCIENCE, HARDWARE & ARCHITECTURE\",\"Score\":null,\"Total\":0}","platform":"Semanticscholar","paperid":null,"PeriodicalName":"Computer Standards & Interfaces","FirstCategoryId":"94","ListUrlMain":"https://www.sciencedirect.com/science/article/pii/S0920548924000758","RegionNum":2,"RegionCategory":"计算机科学","ArticlePicture":[],"TitleCN":null,"AbstractTextCN":null,"PMCID":null,"EPubDate":"","PubModel":"","JCR":"Q1","JCRName":"COMPUTER SCIENCE, HARDWARE & ARCHITECTURE","Score":null,"Total":0}
引用次数: 0

摘要

背景:在过去几年中,神经网络被工业界和研究领域广泛采用。神经网络的里程碑通常是通过扩大计算量来实现的,完全无视相关计算所需的碳足迹。鉴于深度学习的使用越来越多,这种趋势已变得不可持续,如果不尽快解决,可能会对我们的地球环境造成不可逆转的破坏。目标:在本研究中,我们不仅要分析不同节能方法对神经网络的影响,还要分析干预时刻的影响,以及是什么使某些时刻成为最佳时刻。方法:我们在 12 个不同的计算机视觉数据集中训练卷积神经网络,并在每次运行的不同时间段应用有关层冻结、模型量化和提前停止的运行时决策,从而开发了一个新的数据集。然后,我们在收集到的数据上拟合了一个自动回归预测模型,该模型能够预测不同方法在未来历时中所达到的精度和能耗。结果:根据模型的预测,可以节省 56.5% 的能耗,同时通过避免过度拟合,还能提高 2.38% 的验证精度。结论:该预测模型有可能被训练算法用于决定哪些方法适用于该模型,以及在什么时候使用,以便最大限度地实现准确性-能源权衡。
本文章由计算机程序翻译,如有差异,请以英文原文为准。
Energy-efficient neural network training through runtime layer freezing, model quantization, and early stopping

Background:

In the last years, neural networks have been massively adopted by industry and research in a wide variety of contexts. Neural network milestones are generally reached by scaling up computation, completely disregarding the carbon footprint required for the associated computations. This trend has become unsustainable given the ever-growing use of deep learning, and could cause irreversible damage to the environment of our planet if it is not addressed soon.

Objective:

In this study, we aim to analyze not only the effects of different energy saving methods for neural networks but also the effects of the moment of intervention, and what makes certain moments optimal.

Method:

We developed a novel dataset by training convolutional neural networks in 12 different computer vision datasets and applying runtime decisions regarding layer freezing, model quantization and early stopping at different epochs in each run. We then fit an auto-regressive prediction model on the data collected capable to predict the accuracy and energy consumption achieved on future epochs for different methods. The predictions on accuracy and energy are used to estimate the optimal training path.

Results:

Following the predictions of the model can save 56.5% of energy consumed while also increasing validation accuracy by 2.38% by avoiding overfitting.The prediction model developed can predict the validation accuracy with a 8.4% of error, the energy consumed with a 14.3% of error and the trade-off between both with a 8.9% of error.

Conclusions:

This prediction model could potentially be used by the training algorithm to decide which methods apply to the model and at what moment in order to maximize the accuracy-energy trade-off.

求助全文
通过发布文献求助,成功后即可免费获取论文全文。 去求助
来源期刊
Computer Standards & Interfaces
Computer Standards & Interfaces 工程技术-计算机:软件工程
CiteScore
11.90
自引率
16.00%
发文量
67
审稿时长
6 months
期刊介绍: The quality of software, well-defined interfaces (hardware and software), the process of digitalisation, and accepted standards in these fields are essential for building and exploiting complex computing, communication, multimedia and measuring systems. Standards can simplify the design and construction of individual hardware and software components and help to ensure satisfactory interworking. Computer Standards & Interfaces is an international journal dealing specifically with these topics. The journal • Provides information about activities and progress on the definition of computer standards, software quality, interfaces and methods, at national, European and international levels • Publishes critical comments on standards and standards activities • Disseminates user''s experiences and case studies in the application and exploitation of established or emerging standards, interfaces and methods • Offers a forum for discussion on actual projects, standards, interfaces and methods by recognised experts • Stimulates relevant research by providing a specialised refereed medium.
×
引用
GB/T 7714-2015
复制
MLA
复制
APA
复制
导出至
BibTeX EndNote RefMan NoteFirst NoteExpress
×
提示
您的信息不完整,为了账户安全,请先补充。
现在去补充
×
提示
您因"违规操作"
具体请查看互助需知
我知道了
×
提示
确定
请完成安全验证×
copy
已复制链接
快去分享给好友吧!
我知道了
右上角分享
点击右上角分享
0
联系我们:info@booksci.cn Book学术提供免费学术资源搜索服务,方便国内外学者检索中英文文献。致力于提供最便捷和优质的服务体验。 Copyright © 2023 布克学术 All rights reserved.
京ICP备2023020795号-1
ghs 京公网安备 11010802042870号
Book学术文献互助
Book学术文献互助群
群 号:481959085
Book学术官方微信