关于车联网联邦学习中可信信誉管理方法的研究  

Research on Credible Reputation Management Approach for Federated Learning in Internet of Vehicles

作　　者：周灵灵 付宇钏 李长乐[1] ZHOU Lingling;FU Yuchuan;LI Changle(State Key Laboratory of Integrated Services Networks,Xidian University,Xi'an 710071,China)

机构地区：[1]西安电子科技大学空天地一体化综合业务网全国重点实验室,陕西西安710071

出　　处：《移动通信》2023年第10期32-37,57,共7页Mobile Communications

基　　金：国家自然科学基金重点项目“车联网通感算一体化理论与方法”(62231020)。

摘　　要：联邦学习由于其分布式、隐私保护等特点有望应用到车联网中,然而由于缺少相应的本地模型质量验证机制,全局模型容易受到恶意用户的攻击从而导致模型训练的准确率降低。提出一种车联网中分层区块链使能的联邦学习信誉管理架构。首先介绍整个架构的组成以及具体的工作流程,然后设计智能合约为系统提供更加灵活可信的信誉意见共享环境,并开发一种轻量级的区块链共识算法,以提升区块链的运行效率。仿真结果表明所提方法能够筛选出恶意用户,同时保证数据隐私和安全,从而提高FL的准确性。Federated Learning(FL)is expected to be applied to the Internet of Vehicles(loV)due to its distributed and privacy-preserving characteristics.However,due to the lack of the quality verification mechanism for local models,the global model is vulnerable to attacks by malicious users,resulting in reduced model training accuracy.This paper proposes a hierarchical blockchain-enabled reputation management architecture for FL in the IoV.Specifically,the composition of the entire architecture and the specific workflow are introduced,then smart contracts are designed to provide a more flexible and credible reputation sharing environment for the system,and a lightweight blockchain consensus algorithm is developed to improve the operation efficiency of the blockchain.Simulation results show that the proposed method can screen out malicious users while ensuring data privacy and security,thereby improving the accuracy of FL.

关 键 词：联邦学习 车联网 信誉管理 

分 类 号：TN929.5[电子电信—通信与信息系统]