When will federated learning transfer from generalization to personalization?

Wenxuan Liu, Lukun Wang, Jiaming Pei

Keywords: Federated Learning, Generalization, Personalization

TL;DR: We propose that the training transfers from generalization to personalization when the accuracy of the global model reaches a predefined threshold to enhance the performance of federated learning.

Abstract: The timing of personalization refers to determining when to train and update personalized models for the participants in Personalized federated learning. Determining the timing for personalization contributes to improving the overall efficiency of federated learning. We propose that training transfers to personalization when the accuracy of the global model reaches a predefined threshold. Experimental results show that this method can effectively improve the accuracy of personalized models in a non-IID scenario.