One Important Thing To Do Before Federated Training

Yichu Xu, Wenqian Li, Yinchuan Li, yunfeng shao, Yan Pang, De-Chuan Zhan

Keywords: Federated Learning

Abstract: Previous research in Federated learning (FL) have emphasized privacy protection, model optimization, and so on, meanwhile, they overlooked how to choose the appropriate FL algorithm for a new federation with preserving data privacy. In our study, we provide a formal problem formulation for algorithm selection in FL and present a novel approach that involves leveraging trained federations to aid with algorithm selection. Empirical results prove the effectiveness of our method.