federated learning has attracted increasing attention due to the promise of
balancing privacy and large-scale learning; numerous approaches have been
proposed. However, most existing approaches focus on problems with balanced
data, and prediction performance is far from satisfactory fo