Turbo-aggregate: Breaking the quadratic aggregation barrier in secure federated learning

被引:120
作者
So J. [1 ]
Güler B. [2 ]
Avestimehr A.S. [1 ]
机构
[1] The Department of Electrical and Computer Engineering, University of Southern California, Los Angeles, 90089, CA
[2] The Department of Electrical and Computer Engineering, University of California at Riverside, Riverside, 92521, CA
来源
IEEE Journal on Selected Areas in Information Theory | 2021年 / 2卷 / 01期
关键词
Federated learning; Privacy-preserving machine learning; Secure aggregation;
D O I
10.1109/JSAIT.2021.3054610
中图分类号
学科分类号
摘要
Federated learning is a distributed framework for training machine learning models over the data residing at mobile devices, while protecting the privacy of individual users. A major bottleneck in scaling federated learning to a large number of users is the overhead of secure model aggregation across many users. In particular, the overhead of the state-of-the-art protocols for secure model aggregation grows quadratically with the number of users. In this article, we propose the first secure aggregation framework, named Turbo-Aggregate, that in a network with N users achieves a secure aggregation overhead of O(N log N), as opposed to O(N2 ), while tolerating up to a user dropout rate of 50%. Turbo-Aggregate employs a multi-group circular strategy for efficient model aggregation, and leverages additive secret sharing and novel coding techniques for injecting aggregation redundancy in order to handle user dropouts while guaranteeing user privacy. We experimentally demonstrate that Turbo-Aggregate achieves a total running time that grows almost linear in the number of users, and provides up to 40× speedup over the state-of-the-art protocols with up to N = 200 users. Our experiments also demonstrate the impact of model size and bandwidth on the performance of Turbo-Aggregate. © 2021 IEEE.
引用
收藏
页码:479 / 489
页数:10
相关论文
共 55 条
[41]  
Smith V., Chiang C.-K., Sanjabi M., Talwalkar A.S., Federated multi-task learning, Proc. Adv. Neural Inf. Process. Syst., pp. 4424-4434, (2017)
[42]  
Mohri M., Sivek G., Suresh A.T., Agnostic Federated Learning, (2019)
[43]  
Li T., Sanjabi M., Smith V., Fair Resource Allocation in Federated Learning, (2019)
[44]  
Li X., Huang K., Yang W., Wang S., Zhang Z., On the Convergence of Fedavg on Non-IID Data, (2019)
[45]  
Syta E., Et al., Scalable bias-resistant distributed randomness, Proc. IEEE Symp. Security Privacy (SP), pp. 444-460, (2017)
[46]  
Diffie W., Hellman M., New directions in cryptography, IEEE Trans. Inf. Theory, 22, 6, pp. 644-654, (2006)
[47]  
So J., Guler B., Avestimehr A.S., Mohassel P., CodedPrivateML: A Fast and Privacy-Preserving Framework for Distributed Machine Learning, (2019)
[48]  
So J., Guler B., Avestimehr A.S., A scalable approach for privacy-preserving collaborative machine learning, Proc. Adv. Neural Inf. Process. Syst., (2020)
[49]  
Cover T.M., Thomas J.A., Elements of Information Theory (Telecommunications and Signal Processing), (2006)
[50]  
He C., Et al., FedML: A Research Library and Benchmark for Federated Machine Learning, (2020)