An algorithm that combined prospect theory with differential privacy was proposed to address privacy protection and data quality issues in federated learning.From the perspective of maximizing the utility of data holders based on prospect theory,the incentive problem of data holders was transformed into a utility optimization problem,and the optimal reward and punish-ment strategy was found to motivate users to participate in federated learning.An evolutionary game model based on prospect theory was constructed.The evolution trend of the game model in different theoretical application scenarios was analyzed using local stability analysis and numerical simulation.Experimental results show that the proposed method can increase the proportion of users participating in federated training,increase the accuracy of the ultimately shared federated learning model,and reduce the risk of user privacy leakage.
关键词
联邦学习/隐私保护/前景理论/差分隐私/效用优化/最优奖惩策略/演化博弈
Key words
federated learning/privacy protection/prospect theory/differential privacy/utility optimization/optimal reward and punishment strategy/evolutionary game
维普
万方数据