分布式联邦学习结合联盟博弈在物联网中的应用  

Application of Distributed Federated Learning Combined with Coalition Game in the Internet of Things

作　　者：祁中富 张志才 QI Zhongfu;ZHANG Zhicai(College of Physics and Electronic Engineering,Shanxi University,Taiyuan 030006,China;School of Computer Science and Technology,Hainan University,Haikou 570228,China)

机构地区：[1]山西大学物理电子工程学院,山西太原030006 [2]海南大学计算机科学与技术学院,海南海口570228

出　　处：《测试技术学报》2025年第2期230-237,共8页Journal of Test and Measurement Technology

基　　金：山西省基础研究计划自然科学研究面上资助项目(202103021224024);山西省基础研究计划青年科学研究资助项目(202103021223021);山西省重点研发计划资助项目(202202020101004)。

摘　　要：联邦学习作为一种新兴的分布式机器学习范式,在物联网应用中能有效保护数据隐私,但仍存在模型更新效率低和实时性差等问题。为此,提出将分布式联邦学习与联盟博弈相结合的新模型,在该模型中客户端通过权衡形成联盟的收益和成本合作进行联邦学习。由于物联网中设备资源的有限性,提出在联盟中选择领导者协调训练过程。为了引导客户端自适应地形成联盟,同时确保更新模型的精度和效率,设计了一种分布式联盟形成算法。通过联盟合并和拆分操作的不断执行,寻找到最终的联盟分区,最大化合作设备的效用。为了实现联盟成本的公平分配,提出了一种成本分摊机制,维持算法给定结果的稳定性。最后,通过与其他策略的实验对比分析,验证了所提出模型的有效性。As a new distributed machine learning paradigm,federated learning can effectively protect data privacy in Internet of Things(IoT)applications,but it still faces challenges such as low model update effi-ciency and poor real-time performance.To address these issues,a new model combining distributed feder-ated learning and coalition games is proposed,where clients collaborate in federated learning by weighing the benefits and costs of forming a coalition.Given the limited resources of devices in the IoT,a leader is selected within the coalition to coordinate the training process.To guide clients in adaptively forming coali-tions while ensuring the accuracy and efficiency of model updates,a distributed coalition formation algo-rithm was designed.Through continuous execution of coalition mergers and splits,the ultimate coalition partition is achieved to maximize the utility of cooperative devices.To ensure a fair distribution of coali-tion costs,a cost allocation mechanism is proposed to maintain the stability of the algorithm's results.Finally,experimental comparisons with other strategies validate the effectiveness of the proposed model.

关 键 词：联邦学习 模型更新效率 联盟博弈 选择领导者 分布式联盟形成 

分 类 号：TP393[自动化与计算机技术—计算机应用技术]