Federated Learning Incentive Mechanism Based on User Ego-Drive

Federated Learning (FL) is a novel distributed machine learning paradigm developed to overcome data silos and privacy protection issues in the domain of large-scale machine learning. However, in practical applications, users often reject participation in training due to computation and communication costs, and even those who participate in the training lack the internal drive to sustain consistent involvement. To address these challenges, this paper proposes a multi-dimensional Federated Learning incentive mechanism based on user drive (IMUD—Incentive mechanism Based on User Ego-Drive). IMUD designs different incentive schemes based on user drive while considering internal motivation as a crucial factor in traditional incentive mechanisms. The proposed mechanism enhances the model's accuracy by encouraging users to provide better training via a two-layer Stackelberg game model centered on the model's consumers, trainers, and providers. The theoretical analysis of this proposed incentive mechanism confirms individual rationality, incentive compatibility, and other constrained properties. The experiments reveal that IMUD successfully improves users’ participation and enhances model training accuracy while prioritizing reasonable loss and fairness. These results demonstrate the mechanism's significant social benefits.