MPAF: Model Poisoning Attacks to Federated Learning based on Fake Clients

被引:50
作者
Cao, Xiaoyu [1 ]
Gong, Neil Zhenqiang [1 ]
机构
[1] Duke Univ, Durham, NC 27706 USA
来源
2022 IEEE/CVF CONFERENCE ON COMPUTER VISION AND PATTERN RECOGNITION WORKSHOPS, CVPRW 2022 | 2022年
基金
美国国家科学基金会;
关键词
D O I
10.1109/CVPRW56347.2022.00383
中图分类号
TP301 [理论、方法];
学科分类号
081202 ;
摘要
Existing model poisoning attacks to federated learning assume that an attacker has access to a large fraction of compromised genuine clients. However, such assumption is not realistic in production federated learning systems that involve millions of clients. In this work, we propose the first Model Poisoning Attack based on Fake clients called MPAF. Specifically, we assume the attacker injects fake clients to a federated learning system and sends carefully crafted fake local model updates to the cloud server during training, such that the learnt global model has low accuracy for many indiscriminate test inputs. Towards this goal, our attack drags the global model towards an attacker-chosen base model that has low accuracy. Specifically, in each round of federated learning, the fake clients craft fake local model updates that point to the base model and scale them up to amplify their impact before sending them to the cloud server. Our experiments show that MPAF can significantly decrease the test accuracy of the global model, even if classical defenses and norm clipping are adopted, highlighting the need for more advanced defenses.
引用
收藏
页码:3395 / 3403
页数:9
相关论文
共 30 条
  • [1] [Anonymous], 2000, ANDROID X86 RUN ANDR
  • [2] [Anonymous], 2017, Federated learning: Collaborative machine learning without centralized training data.
  • [3] [Anonymous], NOXPLAYER PERFECT AN
  • [4] [Anonymous], Utilization of FATE in Risk Management of Credit in Small and Micro Enterprises
  • [5] [Anonymous], 2021, ACQUIRE VALUED SHOPP
  • [6] [Anonymous], WORLDS 1 CLOUD BASED
  • [7] [Anonymous], Machine Learning Ledger Orchestration For Drug Discovery (MELLODDY)
  • [8] [Anonymous], 2018, ICML
  • [9] Bagdasaryan E, 2020, PR MACH LEARN RES, V108, P2938
  • [10] BHAGOJI AN, 2019, ICML, V97