Delving into the Adversarial Robustness of Federated Learning

In Federated Learning (FL), models are as fragile centrally trained against adversarial examples. However, the robustness of federated learning remains largely unexplored. This paper casts light on challenge learning. To facilitate a better understanding vulnerability existing FL methods, we conduct comprehensive evaluations various attacks and training methods. Moreover, reveal negative impacts induced by directly adopting in FL, which seriously hurts test accuracy, especially non-IID settings. this work, propose novel algorithm called Decision Boundary based Adversarial Training (DBFAT), consists two components (local re-weighting global regularization) to improve both accuracy systems. Extensive experiments multiple datasets demonstrate that DBFAT consistently outperforms other baselines under IID