A Truthful Incentive Scheme Based on Data Forgetting Game for Federated Unlearning

Federated unlearning (FUL) is a compelling method that enables users to eliminate the impact of their data from federated learning (FL) models, thereby ensuring their right to be forgotten while enhancing data privacy. Nevertheless, existing research on FUL primarily concentrates on designing effective unlearning methods, with comparatively little focus on motivating client engagement and mitigating self-serving behaviors. To address the aforementioned challenges, we present a truthful incentive scheme based on data forgetting game for FUL. In particular, we first propose a practical FUL framework where the FL server authorizes clients to selectively revoke portions of their data. Next, to encourage clients to maintain a larger portion of their data during unlearning, we introduce a truthful incentive mechanism grounded in a data forgetting game, which analyzes the FL server's optimal incentive strategy and the corresponding data revocation decisions made by clients. Furthermore, a zero-payment mechanism is developed to curb clients' selfish behavior. Finally, simulation results demonstrate that our approach effectively restrains selfishness and enhances the accuracy of the unlearned models.