A Secure Dynamic Incentive Scheme for Federated Learning

Federated learning (FL) has gained attention as a promising paradigm for collaborative model training across distributed data sources, such as mobile devices and edge servers in Internet of Things (IoT) edge computing scenarios. However, issues such as data privacy security, dynamic node heterogeneity, and different willingness to contribute are still severe. We propose a dynamic incentive scheme for secure federated learning (SDFL) to address these issues. It promotes ongoing participation and cooperation based on participant behavior fluctuations and rewards contributions. In addition, we introduced regulatory solid measures (BCRSA) to reduce potential confrontational behavior while protecting participant privacy. We also propose a Dynamic Consensus Collaborative Strategy Algorithm (DCSA) using Markov Decision Process (MDP) to attract honest participation and ensure fairness. Finally, we proved the robustness and effectiveness of the scheme through rigorous theoretical analysis and simulation experiments.