计算机科学
激励
智能合约
块链
可扩展性
可靠性
计算机安全
Byzantine容错
可信计算
分布式计算
容错
数据库
经济
微观经济学
政治学
法学
作者
Jiuzheng Wang,Ruilin Zhang,Xinyi Li,Hao Yin
标识
DOI:10.1109/trustcom60117.2023.00066
摘要
In the realm of federated learning systems, establishing fair and trustworthy incentive mechanisms stands as a pivotal challenge. Unlike conventional distributed machine learning, federated learning operates within a decentralized client cluster, where participants meticulously assess incentives and costs before opting to engage. This paper introduces FedJudge, a novel Blockchain-based incentive mechanism that ensures trustworthiness throughout the complete lifecycle of federated learning. To address the issue of evaluating clients' contributions to the federated model, we adapt the Shapley value algorithm from game theory, resulting in FedShapley. This framework impartially and credibly quantifies the marginal impact of each client on the federated model's advancement. To ensure objectivity, credibility, and scalability in the FedShapley computation process, we propose FedShapleyPMC—a trusted parallel algorithm harnessing smart contract technology on the blockchain. Furthermore, we implement an automated payment allocation system based on a cryptographic token infrastructure on the blockchain. This implementation guarantees a trustworthy incentive mechanism throughout the federated learning process. Through empirical validation and analysis on authentic datasets, we demonstrate that FedJudge significantly enhances Byzantine fault tolerance while concurrently reducing computation and communication complexities. These advancements are achieved without compromising the robust privacy and security safe-guards.
科研通智能强力驱动
Strongly Powered by AbleSci AI