Design of Federal Learning Incentive Mechanism in Non-IID Data Environment
In the Federated Learning environment,the existence of Non-Independent Identically Distributed(Non-IID)data poses a serious challenge to model performance and user engagement.To address these challenges,this paper proposes a new incentive mechanism based on game theory and Deep Reinforcement Learning,to improve the Federated Learning effect in Non-IID data environment.By designing the payofffunction of the central server and the user,considering the communication cost,computing cost and local model accuracy,the user contribution is measured fairly,and the user participation strategy is optimized by using the game theory model and the Deep Reinforcement Learning algorithm.The experimental results show that the proposed incentive mechanism significantly improves the accuracy of the model and the participation of users,and effectively alleviates the negative impact of Non-IID data distribution on Federated Learning performance,so as to enhance the performance and stability of the whole system.
国家科技期刊平台
NSTL
万方数据
