To protect vehicular privacy and speed up the execution of tasks,federated learning is introduced in the Internet of Vehicles(IoV)where users execute model training locally and upload local models to the base station ...To protect vehicular privacy and speed up the execution of tasks,federated learning is introduced in the Internet of Vehicles(IoV)where users execute model training locally and upload local models to the base station without massive raw data exchange.However,heterogeneous computing and communication resources of vehicles cause straggler effect which weakens the reliability of federated learning.Dropping out vehicles with limited resources confines the training data.As a result,the accuracy and applicability of federated learning models will be reduced.To mitigate the straggler effect and improve performance of federated learning,we propose a reconfigurable intelligent surface(RIS)-assisted federated learning framework to enhance the communication reliability for parameter transmission in the IoV.Furthermore,we optimize the phase shift of RIS to achieve a more reliable communication environment.In addition,we define vehicular competence to measure both vehicular trustworthiness and resources.Based on the vehicular competence,the straggler effect is mitigated where training tasks of computing stragglers are offloaded to surrounding vehicles with high competence.The experiment results verify that our proposed framework can improve the reliability of federated learning in terms of computing and communication in the IoV.展开更多
联邦学习作为一种灵活且可扩展的分布式机器学习方法,在工业物联网(IIoT,industrial Internet of things)中得到了广泛应用,在保护数据隐私的同时,实现低时延、低通信开销和高精度的模型训练。然而,由于工业物联网中边缘设备的计算能力...联邦学习作为一种灵活且可扩展的分布式机器学习方法,在工业物联网(IIoT,industrial Internet of things)中得到了广泛应用,在保护数据隐私的同时,实现低时延、低通信开销和高精度的模型训练。然而,由于工业物联网中边缘设备的计算能力和通信能力的异构性,传统同步联邦学习面临“落后者效应”,即服务器需要等待所有客户端完成本地模型参数上传,显著降低训练效率,难以满足工业物联网对低时延服务的需求。为了解决这一问题并降低设备异构性带来的训练时延,提出了一种基于半同步机制的异构工业联邦学习框架,并在此基础上设计了一种基于训练时延效益评分的客户端选择方案,以提升训练效率。此外,为了提高网络频谱资源的利用率,基于全局训练时延均衡的数学关系,提出了一种自适应设备数量的带宽分配机制,优化被选客户端的模型上传策略。大量仿真结果表明,与基于加权平均的联邦学习(FedAvg,federated averaging)和结合客户端选择的联邦学习(FedCS,federated learning with client selection)等基准方案相比,所提方法在模型准确度、系统时延及频谱利用率等方面均具有显著优势。展开更多
Federated learning(FL)has becoming a prevailing paradigm which enables small-scale devices to collaboratively learn a shared model efficiently and trains a machine learning model without exchanging data.However,though...Federated learning(FL)has becoming a prevailing paradigm which enables small-scale devices to collaboratively learn a shared model efficiently and trains a machine learning model without exchanging data.However,though the original data never leave the local machines in federated learning,possible privacy leakage still exists.To make strong privacy guarantee,in this paper,we incorporate the notion of differential privacy(DP)to study the federated averaging(FedAvg)algorithm.In particular,by adding calibrated gaussian noise,we propose a set of differentially private federated averaging algorithms(DP-FedAvg)under the full and partial participation schemes.We provide tight analysis of the privacy bound by using advanced composition and privacy amplification techniques.We also analyze the convergence bound of DP-FedAvg without having the assumptions:(i)the data are the independent identically distribution(IID),and(ii)all the devices are active.It turns out that the convergence rate is consistent with the one without DP guarantee.The effectiveness of our algorithms is demonstrated by synthetic and real datasets.展开更多
基金supported in part by the Fundamental Research Funds for the Central Universities (2022JBQY004)the Beijing Natural Science Foundation L211013+4 种基金the Basic Research Program under Grant JCKY2022XXXX145the National Natural Science Foundation of China (No. 62221001,62201030)the Science and Technology Research and Development Plan of China Railway Co., Ltd (No. K2022G018)the project of CHN Energy Shuohuang Railway under Grant SHTL-2332the China Postdoctoral Science Foundation 2021TQ0028,2021M700369
文摘To protect vehicular privacy and speed up the execution of tasks,federated learning is introduced in the Internet of Vehicles(IoV)where users execute model training locally and upload local models to the base station without massive raw data exchange.However,heterogeneous computing and communication resources of vehicles cause straggler effect which weakens the reliability of federated learning.Dropping out vehicles with limited resources confines the training data.As a result,the accuracy and applicability of federated learning models will be reduced.To mitigate the straggler effect and improve performance of federated learning,we propose a reconfigurable intelligent surface(RIS)-assisted federated learning framework to enhance the communication reliability for parameter transmission in the IoV.Furthermore,we optimize the phase shift of RIS to achieve a more reliable communication environment.In addition,we define vehicular competence to measure both vehicular trustworthiness and resources.Based on the vehicular competence,the straggler effect is mitigated where training tasks of computing stragglers are offloaded to surrounding vehicles with high competence.The experiment results verify that our proposed framework can improve the reliability of federated learning in terms of computing and communication in the IoV.
文摘联邦学习作为一种灵活且可扩展的分布式机器学习方法,在工业物联网(IIoT,industrial Internet of things)中得到了广泛应用,在保护数据隐私的同时,实现低时延、低通信开销和高精度的模型训练。然而,由于工业物联网中边缘设备的计算能力和通信能力的异构性,传统同步联邦学习面临“落后者效应”,即服务器需要等待所有客户端完成本地模型参数上传,显著降低训练效率,难以满足工业物联网对低时延服务的需求。为了解决这一问题并降低设备异构性带来的训练时延,提出了一种基于半同步机制的异构工业联邦学习框架,并在此基础上设计了一种基于训练时延效益评分的客户端选择方案,以提升训练效率。此外,为了提高网络频谱资源的利用率,基于全局训练时延均衡的数学关系,提出了一种自适应设备数量的带宽分配机制,优化被选客户端的模型上传策略。大量仿真结果表明,与基于加权平均的联邦学习(FedAvg,federated averaging)和结合客户端选择的联邦学习(FedCS,federated learning with client selection)等基准方案相比,所提方法在模型准确度、系统时延及频谱利用率等方面均具有显著优势。
基金Supported by National Natural Science Foundation of China(Grant Nos.12326615,12301384)the Major Key Project of PCL under Grant PCL2024A06。
文摘Federated learning(FL)has becoming a prevailing paradigm which enables small-scale devices to collaboratively learn a shared model efficiently and trains a machine learning model without exchanging data.However,though the original data never leave the local machines in federated learning,possible privacy leakage still exists.To make strong privacy guarantee,in this paper,we incorporate the notion of differential privacy(DP)to study the federated averaging(FedAvg)algorithm.In particular,by adding calibrated gaussian noise,we propose a set of differentially private federated averaging algorithms(DP-FedAvg)under the full and partial participation schemes.We provide tight analysis of the privacy bound by using advanced composition and privacy amplification techniques.We also analyze the convergence bound of DP-FedAvg without having the assumptions:(i)the data are the independent identically distribution(IID),and(ii)all the devices are active.It turns out that the convergence rate is consistent with the one without DP guarantee.The effectiveness of our algorithms is demonstrated by synthetic and real datasets.
