联邦学习安全与隐私保护综述

被引:31
作者
陈兵
成翔
张佳乐
谢袁源
机构
[1] 南京航空航天大学计算机科学与技术学院/人工智能学院
关键词
计算机系统结构; 联邦学习; 模型安全; 隐私保护;
D O I
10.16356/j.1005-2615.2020.05.001
中图分类号
TP309 [安全保密];
学科分类号
081201 ; 0839 ; 1402 ;
摘要
联邦学习是一种新型的分布式学习框架,它允许在多个参与者之间共享训练数据而不会泄露其数据隐私。但是这种新颖的学习机制仍然可能受到来自各种攻击者的前所未有的安全和隐私威胁。本文主要探讨联邦学习在安全和隐私方面面临的挑战。首先,本文介绍了联邦学习的基本概念和威胁模型,有助于理解其面临的攻击。其次,本文总结了由内部恶意实体发起的3种攻击类型,同时分析了联邦学习体系结构的安全漏洞和隐私漏洞。然后从差分隐私、同态密码系统和安全多方聚合等方面研究了目前最先进的防御方案。最后通过对这些解决方案的总结和比较,进一步讨论了该领域未来的发展方向。
引用
收藏
页码:675 / 684
页数:10
相关论文
共 64 条
  • [1] Blockchain and federated learning for privacy-preserved data sharing in industrial IoT. LU Y,HUANG X,DAI Y,et al. IEEE Transactions on Industrial Informatics . 2019
  • [2] Support vector machines under adversarial label contamination. Xiao H,Biggio B,Nelson B,et al. Neurocomputing . 2015
  • [3] A Reflection of Future in History: Introduction to The Alfred North Whitehead Laureate Lecture
    Fei-Yue Wang
    [J]. IEEE/CAAJournalofAutomaticaSinica, 2019, 6 (03) : 609 - 609
  • [4] Editors' declarations of interest
    不详
    [J]. ALIMENTARY PHARMACOLOGY & THERAPEUTICS, 2019, 49 (01) : 6 - 6
  • [5] Ensemble adversarial training:Attacks and defenses. Tramèr F,Kurakin A,Papernot N.et al. https://arxiv.org/pdf/ 1705.07204 . 2017
  • [6] Machine learning on big data: Opportunities and challenges[J] . Lina Zhou,Shimei Pan,Jianwu Wang,Athanasios V. Vasilakos. &nbspNeurocomputing . 2017
  • [7] Situating learning analytics pedagogically: towards an ecological lens[J] . Tan,Koh. &nbspLearning: Research and Practice . 2017 (1)
  • [8] The security of machine learning
    Barreno, Marco
    Nelson, Blaine
    Joseph, Anthony D.
    Tygar, J. D.
    [J]. MACHINE LEARNING, 2010, 81 (02) : 121 - 148
  • [9] Understanding distributed poisoning attack in federated learning. CAO D,CHANG S,LIN Z,et al. Proceeding of the 25th International Conference on Parallel and Distributed Systems(ICPADS) . 2019
  • [10] Towards the science of security and privacy in machine learning. PAPERNOT N,MCDANIEL P,SINHA A,et al. . 2016