Communicational and Computational Efficient Federated Domain Adaptation

The emerging paradigm of Federated Learning enables mobile users to collaboratively train a model without disclosing their privacy-sensitive data. Nevertheless, data collected from different may not be independent and identically distributed. Thus directly applying the trained new user usually leads performance degradation due so-called domain shift. Unsupervised Domain Adaptation is an effective technique mitigate shift transfer knowledge labeled source domains unlabeled target domain. In this article, we design framework that extends with constraints for preserve privacy all domains. As devices have limited computation communication capabilities, set optimization methods significantly enhance our framework’s efficiency, making it more friendly resource-constrained edge devices. Evaluation results on three datasets show has comparable standard centralized training approach, can reduce overheads by up two orders magnitude.