神经机器翻译周期学习率的实证研究

IF 2.3 3区 计算机科学 Q3 COMPUTER SCIENCE, ARTIFICIAL INTELLIGENCE
Weixuan Wang, Choon Meng Lee, Jianfeng Liu, Talha Çolakoğlu, Wei Peng
{"title":"神经机器翻译周期学习率的实证研究","authors":"Weixuan Wang, Choon Meng Lee, Jianfeng Liu, Talha Çolakoğlu, Wei Peng","doi":"10.1017/s135132492200002x","DOIUrl":null,"url":null,"abstract":"\n In training deep learning networks, the optimizer and related learning rate are often used without much thought or with minimal tuning, even though it is crucial in ensuring a fast convergence to a good quality minimum of the loss function that can also generalize well on the test dataset. Drawing inspiration from the successful application of cyclical learning rate policy to computer vision tasks, we explore how cyclical learning rate can be applied to train transformer-based neural networks for neural machine translation. From our carefully designed experiments, we show that the choice of optimizers and the associated cyclical learning rate policy can have a significant impact on the performance. In addition, we establish guidelines when applying cyclical learning rates to neural machine translation tasks.","PeriodicalId":49143,"journal":{"name":"Natural Language Engineering","volume":"88 1","pages":"316-336"},"PeriodicalIF":2.3000,"publicationDate":"2022-02-09","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"","citationCount":"3","resultStr":"{\"title\":\"An empirical study of cyclical learning rate on neural machine translation\",\"authors\":\"Weixuan Wang, Choon Meng Lee, Jianfeng Liu, Talha Çolakoğlu, Wei Peng\",\"doi\":\"10.1017/s135132492200002x\",\"DOIUrl\":null,\"url\":null,\"abstract\":\"\\n In training deep learning networks, the optimizer and related learning rate are often used without much thought or with minimal tuning, even though it is crucial in ensuring a fast convergence to a good quality minimum of the loss function that can also generalize well on the test dataset. Drawing inspiration from the successful application of cyclical learning rate policy to computer vision tasks, we explore how cyclical learning rate can be applied to train transformer-based neural networks for neural machine translation. From our carefully designed experiments, we show that the choice of optimizers and the associated cyclical learning rate policy can have a significant impact on the performance. In addition, we establish guidelines when applying cyclical learning rates to neural machine translation tasks.\",\"PeriodicalId\":49143,\"journal\":{\"name\":\"Natural Language Engineering\",\"volume\":\"88 1\",\"pages\":\"316-336\"},\"PeriodicalIF\":2.3000,\"publicationDate\":\"2022-02-09\",\"publicationTypes\":\"Journal Article\",\"fieldsOfStudy\":null,\"isOpenAccess\":false,\"openAccessPdf\":\"\",\"citationCount\":\"3\",\"resultStr\":null,\"platform\":\"Semanticscholar\",\"paperid\":null,\"PeriodicalName\":\"Natural Language Engineering\",\"FirstCategoryId\":\"94\",\"ListUrlMain\":\"https://doi.org/10.1017/s135132492200002x\",\"RegionNum\":3,\"RegionCategory\":\"计算机科学\",\"ArticlePicture\":[],\"TitleCN\":null,\"AbstractTextCN\":null,\"PMCID\":null,\"EPubDate\":\"\",\"PubModel\":\"\",\"JCR\":\"Q3\",\"JCRName\":\"COMPUTER SCIENCE, ARTIFICIAL INTELLIGENCE\",\"Score\":null,\"Total\":0}","platform":"Semanticscholar","paperid":null,"PeriodicalName":"Natural Language Engineering","FirstCategoryId":"94","ListUrlMain":"https://doi.org/10.1017/s135132492200002x","RegionNum":3,"RegionCategory":"计算机科学","ArticlePicture":[],"TitleCN":null,"AbstractTextCN":null,"PMCID":null,"EPubDate":"","PubModel":"","JCR":"Q3","JCRName":"COMPUTER SCIENCE, ARTIFICIAL INTELLIGENCE","Score":null,"Total":0}
引用次数: 3

摘要

在训练深度学习网络时,优化器和相关的学习率通常没有经过太多的思考或进行最小的调整,即使它对于确保快速收敛到一个高质量的最小损失函数(也可以在测试数据集上很好地泛化)至关重要。从周期学习率策略在计算机视觉任务中的成功应用中获得灵感,我们探索了如何将周期学习率应用于训练基于变压器的神经网络,用于神经机器翻译。从我们精心设计的实验中,我们表明优化器的选择和相关的周期性学习率策略可以对性能产生重大影响。此外,我们在将周期性学习率应用于神经机器翻译任务时建立了指导方针。
本文章由计算机程序翻译,如有差异,请以英文原文为准。
An empirical study of cyclical learning rate on neural machine translation
In training deep learning networks, the optimizer and related learning rate are often used without much thought or with minimal tuning, even though it is crucial in ensuring a fast convergence to a good quality minimum of the loss function that can also generalize well on the test dataset. Drawing inspiration from the successful application of cyclical learning rate policy to computer vision tasks, we explore how cyclical learning rate can be applied to train transformer-based neural networks for neural machine translation. From our carefully designed experiments, we show that the choice of optimizers and the associated cyclical learning rate policy can have a significant impact on the performance. In addition, we establish guidelines when applying cyclical learning rates to neural machine translation tasks.
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
Natural Language Engineering
Natural Language Engineering COMPUTER SCIENCE, ARTIFICIAL INTELLIGENCE-
CiteScore
5.90
自引率
12.00%
发文量
60
审稿时长
>12 weeks
期刊介绍: Natural Language Engineering meets the needs of professionals and researchers working in all areas of computerised language processing, whether from the perspective of theoretical or descriptive linguistics, lexicology, computer science or engineering. Its aim is to bridge the gap between traditional computational linguistics research and the implementation of practical applications with potential real-world use. As well as publishing research articles on a broad range of topics - from text analysis, machine translation, information retrieval and speech analysis and generation to integrated systems and multi modal interfaces - it also publishes special issues on specific areas and technologies within these topics, an industry watch column and book reviews.
×
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