federated learning (FL) has gained significant attention recently as a
privacy-enhancing tool to jointly train a machine learning model by multiple
participants. The prior work on FL has mostly studied how to protect label
privacy during model training. However, model evaluation in FL