Abstract: Federated learning (FL) can complete collaborative model training without transferring local data, which can greatly improve the training efficiency. However, FL is susceptible data and ...