共识机制在联邦学习中的研究现状
Research Status of Consensus Mechanisms in Federated Learning
摘要
Abstract
The rapid development of federated learning(FL)has ushered in new opportunities for collaborative training on distributed data.Blockchain offers promising solutions to key challenges in FL,such as centralized trust issues,data privacy protection,system security,and communication overhead optimization,thereby becoming an integral component of federated learning.The consensus mechanism,as a critical element of blockchain within FL,enables the verification of model updates,facilitates decentralized aggregation,and utilizes incentives to ensure active participation.Given the varia-tions in consensus mechanisms employed for FL,this paper categorizes and compares existing consensus schemes in FL from four perspectives:classical consensus,performance-based consensus,protocol consensus,and collaborative consensus.Classical consensus prioritizes block generation and fundamental consistency guarantees.Performance-based consensus adjusts participation levels based on nodes'local training performance.Protocol consensus designs task-specific strategies tailored to FL objectives,enabling customized and efficient aggregation.Collaborative consensus supports cross-chain ver-ification and consistency maintenance in multi-chain and multi-domain environments.By comparing representative algo-rithms,advantages/disadvantages,and applicable environments of different FL consensus types,this paper identifies cur-rent issues,including difficulties in embedding privacy,high communication burdens,and poor adaptation to heteroge-neous environments.Finally,this paper proposes future optimization directions such as mechanism fusion,lightweight de-sign,and privacy integration,providing theoretical support and methodological references for the deeper integration of consensus mechanisms and federated learning.关键词
共识机制/区块链/联邦学习Key words
consensus mechanism/blockchain/federated learning分类
计算机与自动化引用本文复制引用
刘义,吴世伟,蒋澄杰,董慧婷,吴银淼,管新如,蒋胜,张磊..共识机制在联邦学习中的研究现状[J].计算机科学与探索,2025,19(11):2913-2934,22.基金项目
黑龙江省省属本科高校基本科研业务费科研项目(2024-KYYWF-0564) （2024-KYYWF-0564）
佳木斯大学国家基金培育项目(JMSUGPZR2022-014) （JMSUGPZR2022-014）
黑龙江省自主智能与信息处理重点实验室开放课题(ZZXC202302) （ZZXC202302）
佳木斯大学"东极"学术团队项目(DJXSTD202417) （DJXSTD202417）
黑龙江省省属本科高校优秀青年教师基础研究支持计划(YQJH2024239) （YQJH2024239）
黑龙江省外国专家项目(G2024020) （G2024020）
佳木斯大学博士专项科研启动项目(JMSUBZ2024-07) （JMSUBZ2024-07）
黑龙江省自然科学基金联合基金培育项目(PL2024F002). This work was supported by the Basic Research Business Expenses of Provincial Universities in Heilongjiang Province(2024-KYYWF-0564),the National Fund Cultivation Project of Jiamusi University(JMSUGPZR2022-014),the Open Project of Heilongjiang Province Key Laboratory of Autonomous Intelligence and Information Processing(ZZXC202302),the Project of"Dongji"Academic Team of Jiamusi University(DJXSTD202417),the Support Program for Basic Research of Outstanding Young Teachers in Provincial Universities of Heilongjiang Province(YQJH2024239),the Foreign Experts Project of Heilongjiang Province(G2024020),the Doctoral Special Research Start-up Project of Jiamusi University(JMSUBZ2024-07),and the Cultivation Project of the Joint Fund of the Natural Science Founda-tion of Heilongjiang Province(PL2024F002). （PL2024F002）
