A hierarchical blockchain-enabled distributed federated learning system with model contribution based rewarding  

作　　者：Haibo Wang Hongwei Gao Teng Ma Chong Li Tao Jing 

机构地区：[1]The Research Institute of Broadband Wireless Mobile Communication,Beijing Jiaotong University,Beijing,100044,China [2]The Department of Electrical Engineering,Columbia University,New York,10027,United States

出　　处：《Digital Communications and Networks》2025年第1期35-42,共8页数字通信与网络(英文版)

摘　　要：Distributed Federated Learning(DFL)technology enables participants to cooperatively train a shared model while preserving the privacy of their local datasets,making it a desirable solution for decentralized and privacy-preserving Web3 scenarios.However,DFL faces incentive and security challenges in the decentralized framework.To address these issues,this paper presents a Hierarchical Blockchain-enabled DFL(HBDFL)system,which provides a generic solution framework for the DFL-related applications.The proposed system consists of four major components,including a model contribution-based reward mechanism,a Proof of Elapsed Time and Accuracy(PoETA)consensus algorithm,a Distributed Reputation-based Verification Mechanism(DRTM)and an Accuracy-Dependent Throughput Management(ADTM)mechanism.The model contribution-based rewarding mechanism incentivizes network nodes to train models with their local datasets,while the PoETA consensus algorithm optimizes the tradeoff between the shared model accuracy and system throughput.The DRTM improves the system efficiency in consensus,and the ADTM mechanism guarantees that the throughput performance remains within a predefined range while improving the shared model accuracy.The performance of the proposed HBDFL system is evaluated by numerical simulations,with the results showing that the system improves the accuracy of the shared model while maintaining high throughput and ensuring security.

关 键 词：Blockchain Federated learning Consensus scheme Accuracy dependent throughput management 

分 类 号：TN9[电子电信—信息与通信工程]