分层边缘计算中利用联邦学习的差分隐私保护  

Differential Privacy Protection Using Federated Learning in Layered Edge Computing

作　　者：董少华 马新春 樊小超[1] DONG Shao-hua;MA Xin-chun;FAN Xiao-chao(School of Computer Science and Technology,Xinjiang Normal University,Urumqi 830054,China;Xinjiang Electronic Research Institute,Urumqi 830013,China)

机构地区：[1]新疆师范大学计算机科学技术学院,新疆乌鲁木齐830054 [2]新疆电子研究所,新疆乌鲁木齐830013

出　　处：《计算机技术与发展》2025年第4期53-58,共6页Computer Technology and Development

基　　金：新疆维吾尔自治区重点研发计划项目(2022B01008);天山英才培养计划项目(2023TSYCJCO039)。

摘　　要：分层边缘计算通过在资源受限的边缘设备和云服务器上协作运行,旨在最小化推理延迟和保护数据隐私。然而,即使来自边缘设备的原始输入数据没有直接暴露于云中,针对边缘数据的最先进的攻击仍然能够从暴露的本地模型的中间输出中重建原始的私有数据,从而引入严重的隐私风险。为了克服这些限制,提出了一种新的算法,该算法合并了边缘阶段和云阶段的模型,提出了确定最佳聚合时间框架的定性指令,以减少计算和通信费用。通过在客户端和边缘服务器级别实现局部差分隐私,增强了本地模型参数更新时的隐私。在CIFAR-10和MNIST数据集上的实验表明,该算法在训练精度、训练时间和通信-计算权衡方面优于标准的联邦学习方法。且该算法为分层边缘计算的挑战提供了一个很有前途的解决方案,使内容交付更快、移动服务质量更高。Layered edge computing,by operating collaboratively on resource-constrained edge devices and cloud servers,aims to minimize inference latency and protect data privacy.However,even if the raw input data from edge devices is not directly exposed to the cloud,sophisticated attacks targeted at edge data can still reconstruct the original private data from exposed local model outputs,introducing serious privacy risks.To overcome these limitations,a new algorithm is proposed that amalgamates models from both the edge and cloud phases,presenting qualitative directives for determining optimal aggregation time frames to reduce computational and communication costs.Local differential privacy is implemented at the client and edge server levels to enhance privacy during local model parameter updates.Experiments on the CIFAR-10 and MNIST datasets demonstrate that the proposed algorithm outperforms standard FL methods in terms of training accuracy,training time,and communication-computation trade-offs.It offers a promising solution to the challenges of layered edge computing,enabling faster content delivery and higher mobile services quality.

关 键 词：分层边缘计算 联邦学习 差分隐私 边缘协作 数据隐私 

分 类 号：TP391.1[自动化与计算机技术—计算机应用技术]