To address the key issue in federated learning the resource heterogeneity and data heterogeneity of each client, this paper proposes a resource-efficient client selection method for clustered collaborative federated learning. Firstly, where each client is grouped according to its computational power, the average accuracy of each group of clients is used as an indirect metric to select clients in the same group in each round of training. Secondly, the clients are clustered according to the model similarity of each client within each group, and the clients in different clusters within each group are selected. Evaluate the performance of the method proposed in this paper on real datasets, the experimental results show that this method can reduce the global training time, obtain faster and smoother convergence, and achieve a good balance between training efficiency and global model accuracy.
近些年来,现代移动和物联网设备每天都在产生大量数据,由于硬件计算能力的提高和大量数据的产生,人工智能技术飞速发展。在传统的集中式机器学习中,需要把分布在各个终端上的数据集中在中心节点上进行处理,若要训练性能较好的模型,则需以大量数据为基础。然而,除了少数大型组织或政府部门外,大多数个人或公司只有少量数据,并且集中收集所有的数据也会产生高昂的开销。同时,人们对数据安全和隐私保护愈发关注,如《通用数据保护条例》(General data protection regulation,GDPR)[1]禁止将数据集中到中心节点,这使得传统的集中式机器学习难以应用于分散数据的收集和处理。此外,不同企业甚至同一企业的不同部门也可能出于数据隐私目的不愿意共享数据,从而形成“数据孤岛”。为了解决这些问题,谷歌提出了联邦学习[2]这一概念。联邦学习要求服务器将随机初始化的全局模型下发到各个客户端设备,客户端设备不需要共享数据,直接使用本地数据更新模型后,将新的本地模型上传回服务器,服务器聚合收到的所有客户端的本地模型,形成新的全局模型,将其再次下发到全部客户端。通过联邦学习机制,可以在保障不泄露用户数据隐私和不影响数据规范的前提下,实现数据的合理利用。
TankardC. What the GDPR means for businesses[J].Network Security, 2016,2016(6): 5-8.
[2]
McMahanH B, McoreE, RamageD, et al. Communication-efficient learning of deep networks from decentralized data[C]∥Proceeding of the 20th Tnternational Conference on Artificial Intelligence and Statistics,Ft. Lauderdale,USA, 2017.
[3]
AbadiM, ChuA, GoodfellowL, et al. Deep learning with differential privacy[C]∥Proceedings of the 2016 ACM SIGSAC Conference on Computer and Communications Security,Vienna, Austria, 2016:308-318.
[4]
BonawitzK, IvanowV, KreuterB, et al. Practical secure aggregation for privacy-preserving machine learning[C]∥Proceedings of the 2017 ACM SIGSAC Conference on Computer and Communications Security,Dallas, USA, 2017: 1175-1191.
[5]
McMahanH B, RamageD, KunalT, et al. Learning differentially private recurrent language models[J/OL].[2022-11-22]. 2017.
[6]
ChaiZ, AliA, ZawadS, et al. Tifl: A tier-based federated learning system[C]∥Proceedings of the 29th International Symposium on High-Performance Parallel and Distributed Computing,Stockholm, Sweden, 2020: 125-136.
[7]
NishioT, YonetaniR. Client selection for federated learning with heterogeneous resources in mobile edge[C]∥IEEE International Conference on Communications(ICC), Piscataway, USA, 2019:1-7.
[8]
KarimireddyS P, KaleS, MohriM, et al. SCAFFOLD: Stochastic controlled averaging for federated learning[C]∥International Conference on Machine Learning,Online, 2020:5132-5143.
[9]
FraboniY, VidalR, KamenlL, et al. Clustered sampling: Low-variance and improved representativity for clients selection in federated learning[C]∥Preceeding of the 38th International Conference on Machine Learning, Online, 2021: 3407-3416.
[10]
LiT, SahuA K, ZaheerM, et al. Federated optimization in heterogeneous networks[J]. Proceedings of Machine Learning and Systems, 2020, 2: 429-450.
[11]
ShuJ G, ZhongW Z, ZhouY, et al. FLAS: Computation and communication efficient federated learning via adaptive sampling[J]. IEEE Transactions on Network Science and Engineering,2021,9(4): 2003-2014.
[12]
BonawitzK, EichnerH, GrieskampW, et al. Towards federated learning at scale: System design[J]. Proceedings of Machine Learning and Systems, 2019, 1: 374-388.
[13]
WardJ H. Hierarchical grouping to optimize an objective function[J]. Journal of the American Statistical Association,1963, 58(301): 236-244.
[14]
SattlerF, MüllerK R, SamekW. Clustered federated learning: Model-agnostic distributed multitask optimization under privacy constraints[J]. IEEE Transactions on Neural Networks and Learning Systems, 2020, 32(8): 3710-3722.
[15]
LeCunY, BottouL, BengioY, et al. Gradient-based learning applied to document recognition[J]. Proceedings of the IEEE, 1998, 86(11): 2278-2324.
[16]
Krizhevsk, HintonG. Learning multiple layers of features from tiny images[R]. Toronto: University of Toronto, 2009.
[17]
HsuT M H, QiH, BrownM. Measuring the effects of non-identical data distribution for federated visual classification[J/OL].[2022-11-22]. 2019.