基于博弈论优化的高效联邦学习方案  

Efficient Federated Learning Scheme Based on Game Theory Optimization

作　　者：周全兴 李秋贤 丁红发[2] 樊玫玫 ZHOU Quanxing;LI Qiuxian;DING Hongfa;FAN Meimei(School of Big Data Engineering,Kaili University,Kaili,Guizhou 556011,China;School of Information,Guizhou University of Finance and Economics,Guiyang 550025,China;School of Mathematics and Statistics,Guizhou University,Guiyang 550025,China)

机构地区：[1]凯里学院大数据工程学院,贵州凯里556011 [2]贵州财经大学信息学院,贵阳550025 [3]贵州大学数学与统计学院,贵阳550025

出　　处：《计算机工程》2022年第8期144-151,159,共9页Computer Engineering

基　　金：国家自然科学基金(61772008,62002080);贵州省教育厅自然科学研究项目(黔教合KY字[2020179],[2020180],[2021140]);凯里学院做特市(州)高校专项计划项目“基于区块链的黔东南从江香猪溯源体系博弈演化技术研究”;贵州省科技重大专项计划(20183001);贵州财经大学校级科研课题(2020XYB02)。

摘　　要：随着网络信息技术与互联网的发展,数据的隐私与安全问题亟待解决,联邦学习作为一种新型的分布式隐私保护机器学习技术应运而生。针对在联邦学习过程中存在个人数据信息泄露的隐私安全问题,结合Micali-Rabin随机向量表示技术,基于博弈论提出一种具有隐私保护的高效联邦学习方案。根据博弈论激励机制,构建联邦学习博弈模型,通过设置合适的效用函数和激励机制保证参与者的合理行为偏好,同时结合Micali-Rabin随机向量表示技术设计高效联邦学习方案。基于Pedersen承诺机制实现高效联邦学习的隐私保护,以保证联邦学习各参与者的利益和数据隐私,并且全局达到帕累托最优状态。在数字分类数据集上的实验结果表明,该方案不仅提高联邦学习的通信效率,而且在通信开销和数据精确度之间实现平衡。With the continuous development of network information technology and Internet technology,data privacy and security issues need to be addressed urgently.Federated learning has emerged as a new distributed privacy protection machine learning framework.This study proposes an efficient federated learning scheme with privacy protection based on game theory and Micali-Rabin random vector representation technology to address privacy and security issues,such as personal data information leakage in the federated learning process.It uses game theory to design a federated learning game theory model,sets appropriate utility functions and incentive mechanisms to ensure participants’reasonable behavioral preferences,and combines Micali-Rabin random vector representation technology to construct an efficient federated learning scheme.Furthermore,it integrates the Pedersen commitment mechanism to realize the privacy protection of efficient federated learning to ensure the security and privacy of each participant in the federated learning and the interests of each participant,and global achieve Pareto optimal state.The experimental result on the digital classification data set shows that the scheme not only improves the communication efficiency of federated learning but also achieves a balance between communication overhead and data accuracy.

关 键 词：联邦学习 博弈论 帕累托最优 隐私保护 Micali-Rabin随机向量表示技术 

分 类 号：TP309[自动化与计算机技术—计算机系统结构]