DP-FMeta: A Differentially Private Federated Optimization Framework Based on Meta Prediction

Federated learning as an important research direction in the field of smart devices, but the current research suffers from high communication cost, slow model convergence and data leakage. Aiming at the problem of model training that cannot be carried out on resource-constrained devices with higher computational requirements during model training, a differential federation optimisation algorithm based on federated prediction(DP-FMeta) is proposed, and a federated prediction header and meta-learning module are designed, which not only significantly improves the expressive ability of the model, but also achieves collaborative knowledge migration across devices as well as solves the problem of reducing the computational cost of the model. Meanwhile, when facing unprocessed data, the model parameters can be quickly adjusted to ensure the accuracy and robustness of the model.In this paper, we have conducted a large number of experiments on several publicly available datasets, which confirm that the proposed method in this paper can achieve better performance than other methods, and validate the effectiveness of the proposed method.