An Adaptive Differential Privacy Scheme for Council Mechanism in Decentralized Federated Learning

In the context of decentralized federated learning, achieving consensus and protecting data privacy are key challenges. In this environment, there is no central server, so a mechanism is needed to ensure that model updates from different participants can be securely and effectively aggregated. In this paper, we propose a privacy-preserving serverless federated learning framework—Adaptive Differential Privacy-based Committee Mechanism in Federated Learning (ADPCMFL), which aimed at addressing this challenge and achieving committee consensus. This scheme can dynamically adjust the degree of privacy protection based on the characteristics of data and learning needs, prevent inference attacks and Byzantine attacks, thereby improving the effectiveness and accuracy of learning while ensuring security and privacy. Finally, the effectiveness and superiority of the proposed scheme in achieving committee consensus were verified through experiments and data analysis. Compared with existing methods, it can better balance privacy protection and learning performance, providing new ideas and methods for privacy protection and consensus building in decentralized federated learning.