作者投稿和查稿 主编审稿 专家审稿 编委审稿 远程编辑

计算机工程

• •    

联邦学习及其安全与隐私保护研究综述

  • 发布日期:2023-11-28

A Review of Federated Learning and its Security and Privacy Protection

  • Published:2023-11-28

摘要: 联邦学习是一种新兴的分布式机器学习技术,其无需对数据进行收集,只需将数据留在本地就能通过各方协作来训练一个共有模型,解决了传统机器学习中数据难以采集和数据隐私安全问题,随着该技术的应用和发展,研究发现联邦学习中仍可能受到各类攻击,为确保联邦学习的足够安全,研究联邦学习中的攻击方式和相应的隐私保护技术显得尤为重要。首先对联邦学习的相关背景知识进行了介绍,随后对联邦学习的定义进行了简要介绍,总结概述了联邦学习的发展历程及其分类,接着介绍了联邦学习安全三要素,从基于来源和基于安全三要素两个角度分类概述了联邦学习中的安全问题,并综述了其研究进展,而后对隐私保护技术进行了分类,结合相关研究应用具体综述了联邦学习中安全多方计算、同态加密、差分隐私和可信执行环境四种常用隐私保护技术,最后对联邦学习的未来研究方向进行了展望。

Abstract: ederated learning is an emerging distributed machine learning technology. It does not need to collect data, but can train a common model through the cooperation of all parties, which solves the problems of difficult data collection and data privacy security in traditional machine learning. With the application and development of this technology, the research finds that federated learning may still be subject to various attacks. In order to ensure enough security of federated learning, it is very important to study the attack mode and the corresponding privacy protection technology in federated learning. First of all, it introduces the relevant background and knowledge of federated learning, and then gives a brief introduction to the definition of federated learning, and summarizes the development process and Classification of federated learning, then introduces the three elements of federated learning security, from the perspective of sourcing-based and security-based three elements of the classification of security issues in federated learning, It also summarizes its research progress, and then classifies the privacy protection technology, combined with relevant research and application, specifically reviews the secure multi-party computing homomorphic encryption in federated learning Differential privacy and trusted execution environment are four common privacy protection technologies. Finally, the future research direction of federated learning is prospected.