Federated Learning in Non-IID Settings Aided by Differentially Private Synthetic Data

Huancheng Chen, Haris Vikalo; Proceedings of the IEEE/CVF Conference on Computer Vision and Pattern Recognition (CVPR) Workshops, 2023, pp. 5027-5036

Federated learning (FL) is a privacy-promoting framework that enables potentially large number of clients to collaboratively train machine learning models. In an FL system, a server coordinates the collaboration by collecting and aggregating clients' model updates while the clients' data remains local and private. A major challenge in federated learning arises when the local data is non-iid -- the setting in which performance of the learned global model may deteriorate significantly compared to the scenario where the data is identically distributed across the clients. In this paper we propose FedDPMS (Federated Differentially Private Means Sharing), an FL algorithm in which clients augment local datasets with data synthesized using differentially private information collected and communicated by a trusted server. In particular, the server matches the pairs of clients having complementary local datasets and facilitates differentially-private sharing of the means of latent data representations; the clients then deploy variational auto-encoders to enrich their datasets and thus ameliorate the effects of non-iid data distribution. Our experiments on deep image classification tasks demonstrate that FedDPMS outperforms competing state-of-the-art FL methods specifically developed to address the challenge of federated learning on non-iid data.