Abstract: The fusion of federated learning and differential privacy can provide more comprehensive and rigorous privacy protection, thus attracting extensive interests from both academia and industry.