无线网络中保证隐私的异步联邦学习系统资源优化配置
Privacy-Preserving Resource Allocation for Asynchronous Federated Learning in wireless networks
周折耳 1陈小静 1张舜卿 1孙彦赞1
作者信息
- 1. 上海大学通信与信息工程学院,上海 200444
- 折叠
摘要
异步联邦学习(Asynchronous Federated Learning,AFL)因其高效性已成为传统同步联邦学习(Synchronous Federated Learning,SFL)的解决方案.然而,在无线场景中,AFL仍面临有限的通信、计算资源,以及安全威胁等挑战.提出了一种新的结合Transformer编码器的近端策略优化(Proximal Policy Optimization)双阶段算法框架,该框架联合优化了AFL系统的学习时延、能耗和模型精度,并通过参与设备协作干扰的方式保证物理层安全.大量仿真验证,当目标准确率为 0.9 时,所提方案相较于基准方案降低了 74.2%的训练时延与能耗.
Abstract
Asynchronous federated learning(AFL)has become a solution to the inefficiency of synchronous federated learning(SFL).However,AFL still faces challenges such as limited communication and computational resources,as well as security threats in wireless networks.This paper proposes a new two-stage proximal policy optimization algorithm framework that combines Transformer encoders.The framework jointly optimizes the learning latency,energy consumption,and model accuracy while ensuring physical layer security through collaborative jamming by devices.Extensive simulation results show that the proposed approach can reduce training latency and energy consumption by 74.2%compared to baseline when the required test accuracy is 0.9.
关键词
异步联邦学习/物理层安全/Transformer/近端策略优化Key words
asynchronous federated learning/physical layer security/Transformer/proximal policy optimization引用本文复制引用
出版年
2024
NETL
NSTL
万方数据