引用本文
  • 王鹏举,卢江虎,刘博超,葛仕明.资源受限场景中的联邦学习技术综述[J].信息安全学报,已采用    [点击复制]
  • Wang Pengju,Lu Jianghu,Liu Bochao,Ge Shiming.Federated Learning in Resource Constrained Scenarios: A Comprehensive Survey[J].Journal of Cyber Security,Accept   [点击复制]
【打印本页】 【在线阅读全文】【下载PDF全文】 查看/发表评论下载PDF阅读器关闭

过刊浏览    高级检索

本文已被:浏览 76次   下载 0  
资源受限场景中的联邦学习技术综述
王鹏举, 卢江虎, 刘博超, 葛仕明
0
(中国科学院信息工程研究所)
摘要:
随着数字经济的快速发展,数据安全威胁日益严峻,数据安全已成为数字经济时代最紧迫的安全问题。联邦学习作为一种新兴的分布式机器学习框架,在实现数据安全和隐私保护的前提下,聚合多方数据资源,协同构建联合模型,为打破数据孤岛现象和实现数据安全融合提供了一种行之有效的方案,为数字经济发展夯实了安全基础,受到了学术界和工业界的广泛关注。然而,在实际的应用部署中,联邦学习面临着资源受限场景所带来的严峻挑战,资源受限场景中存在计算设备、通信网络和建模数据等一系列资源受限问题,这些问题严重地限制了联邦学习的应用和发展。因此,有必要从资源受限的角度来研究联邦学习,着力解决资源受限场景中的突出问题,以实现高效地部署联邦学习。本文主要探讨了资源受限场景中部署联邦学习的实际解决方案。首先,介绍了数字经济的发展现状和联邦学习的背景知识;其次,讨论了资源受限场景中的联邦学习所面临的问题与挑战;然后,对资源受限场景中的联邦学习的研究现状展开了系统深入地调研,分别从架构高效、通信高效、计算高效和异构融合等四个方面对比分析了典型的联邦学习;最后,对资源受限场景中的联邦学习进行了总结与展望。
关键词:  联邦学习  资源受限  架构高效  通信高效  计算高效  异构融合
DOI:
投稿时间:2022-03-09修订日期:2022-06-19
基金项目:北京市自然科学基金
Federated Learning in Resource Constrained Scenarios: A Comprehensive Survey
Wang Pengju, Lu Jianghu, Liu Bochao, Ge Shiming
(Institute of Information Engineering,Chinese Academy of Sciences)
Abstract:
With the rapid development of the digital economy, the threats to data security are becoming serious by degrees, so data security has become the most urgent security issue in the current digital economy era. As an emerging distributed machine learning framework, federated learning aggregates data resources from multiple parties and realizes the collaborative construction of a federated model under the premise of data security and privacy protection. It provides an effective solution to break the phenomenon of isolated data islands and realize data security fusion, in addition to consolidating the security foundation for the development of the digital economy. As a result, it has received extensive attention from both academia and industry. However, in practical application deployment, it shows that federated learning faces severe challenges brought by resource constrained scenarios. In resource constrained scenarios, there are a series of resource constrained problems such as computing device, communication network and modeling data, which seriously restrict the application and development of federated learning. Therefore, it is necessary to explore federated learning from the perspective of resource constrained scenarios, and it is urgent to solve the outstanding problems in these scenarios, in order to achieve efficient deployment of federated learning. In this survey paper, we focus on practical solutions for deploying federated learning in resource constrained scenarios. Firstly, we introduce the development status of the digital economy and the background knowledge of federated learning. Secondly, we discuss the problems and challenges of federated learning in resource constrained scenarios. Thirdly, we conduct a systematic and in-depth investigation into the current research status of federated learning, and analyze a number of typical federated learning technologies in terms of architecture efficient, communication efficient, computation efficient, and heterogeneous fusion. Finally, we make a summary and outlook for the development trend of federated learning in resource constrained scenarios.
Key words:  federated learning  resource constrained  architecture efficient  communication efficient  computation efficient  heterogeneous fusion