Multi-round decentralized dataset distillation with federated learning for Low Earth Orbit satellite communication

IF 6.2 2区 计算机科学 Q1 COMPUTER SCIENCE, THEORY & METHODS
Jianlong Xu, Mengqing Jin, Jinze Xiao, Dianming Lin, Yuelong Liu
{"title":"Multi-round decentralized dataset distillation with federated learning for Low Earth Orbit satellite communication","authors":"Jianlong Xu,&nbsp;Mengqing Jin,&nbsp;Jinze Xiao,&nbsp;Dianming Lin,&nbsp;Yuelong Liu","doi":"10.1016/j.future.2024.107570","DOIUrl":null,"url":null,"abstract":"<div><div>Satellite communication and Low Earth Orbit (LEO) satellites are important components of the 6G network, widely used for Earth observation tasks due to their low cost and short return period, making them a key technology for 6G network connectivity. Due to limitations in satellite system technology and downlink bandwidth, it is not feasible to download all high-resolution image information to ground stations. Even in existing federated learning (FL) methods, sharing well-trained parts of the model can still bottleneck with increasing model size. To address these challenges, we propose a new federated learning framework (FL-M3D) for LEO satellite communication that employs multi-round decentralized dataset distillation techniques. It allows satellites to independently extract local datasets and transmit them to ground stations instead of exchanging model parameters. Communication costs depend only on the size of the synthesized dataset and do not increase with larger models. However, the heterogeneity of satellite datasets can lead to sample ambiguity and decreased model convergence speed. Therefore, we propose distilling the datasets to mitigate the negative effects of data heterogeneity. Through experiments using real-world image datasets, FL-M3D reduces communication volume in simulated satellite networks by approximately 49.84% and achieves improved model performance.</div></div>","PeriodicalId":55132,"journal":{"name":"Future Generation Computer Systems-The International Journal of Escience","volume":"164 ","pages":"Article 107570"},"PeriodicalIF":6.2000,"publicationDate":"2024-10-24","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"","citationCount":"0","resultStr":null,"platform":"Semanticscholar","paperid":null,"PeriodicalName":"Future Generation Computer Systems-The International Journal of Escience","FirstCategoryId":"94","ListUrlMain":"https://www.sciencedirect.com/science/article/pii/S0167739X2400534X","RegionNum":2,"RegionCategory":"计算机科学","ArticlePicture":[],"TitleCN":null,"AbstractTextCN":null,"PMCID":null,"EPubDate":"","PubModel":"","JCR":"Q1","JCRName":"COMPUTER SCIENCE, THEORY & METHODS","Score":null,"Total":0}
引用次数: 0

Abstract

Satellite communication and Low Earth Orbit (LEO) satellites are important components of the 6G network, widely used for Earth observation tasks due to their low cost and short return period, making them a key technology for 6G network connectivity. Due to limitations in satellite system technology and downlink bandwidth, it is not feasible to download all high-resolution image information to ground stations. Even in existing federated learning (FL) methods, sharing well-trained parts of the model can still bottleneck with increasing model size. To address these challenges, we propose a new federated learning framework (FL-M3D) for LEO satellite communication that employs multi-round decentralized dataset distillation techniques. It allows satellites to independently extract local datasets and transmit them to ground stations instead of exchanging model parameters. Communication costs depend only on the size of the synthesized dataset and do not increase with larger models. However, the heterogeneity of satellite datasets can lead to sample ambiguity and decreased model convergence speed. Therefore, we propose distilling the datasets to mitigate the negative effects of data heterogeneity. Through experiments using real-world image datasets, FL-M3D reduces communication volume in simulated satellite networks by approximately 49.84% and achieves improved model performance.
利用联合学习为低地球轨道卫星通信提供多轮分散式数据集提炼服务
卫星通信和低地球轨道(LEO)卫星是 6G 网络的重要组成部分,因其成本低、返回周期短而被广泛用于地球观测任务,成为 6G 网络连接的关键技术。由于卫星系统技术和下行带宽的限制,将所有高分辨率图像信息下载到地面站是不可行的。即使在现有的联合学习(FL)方法中,共享模型中训练有素的部分也会随着模型规模的增大而出现瓶颈。为了应对这些挑战,我们为低地轨道卫星通信提出了一种新的联合学习框架(FL-M3D),它采用了多轮分散数据集提炼技术。它允许卫星独立提取本地数据集并将其传输到地面站,而不是交换模型参数。通信成本仅取决于合成数据集的大小,不会随着模型的增大而增加。然而,卫星数据集的异质性会导致样本模糊和模型收敛速度下降。因此,我们建议对数据集进行提炼,以减轻数据异质性的负面影响。通过使用真实世界图像数据集进行实验,FL-M3D 将模拟卫星网络中的通信量减少了约 49.84%,并提高了模型性能。
本文章由计算机程序翻译,如有差异,请以英文原文为准。
求助全文
约1分钟内获得全文 求助全文
来源期刊
CiteScore
19.90
自引率
2.70%
发文量
376
审稿时长
10.6 months
期刊介绍: Computing infrastructures and systems are constantly evolving, resulting in increasingly complex and collaborative scientific applications. To cope with these advancements, there is a growing need for collaborative tools that can effectively map, control, and execute these applications. Furthermore, with the explosion of Big Data, there is a requirement for innovative methods and infrastructures to collect, analyze, and derive meaningful insights from the vast amount of data generated. This necessitates the integration of computational and storage capabilities, databases, sensors, and human collaboration. Future Generation Computer Systems aims to pioneer advancements in distributed systems, collaborative environments, high-performance computing, and Big Data analytics. It strives to stay at the forefront of developments in grids, clouds, and the Internet of Things (IoT) to effectively address the challenges posed by these wide-area, fully distributed sensing and computing systems.
×
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