Metric-privacy-inspired noise calibration in federated learning: Improving convergence and preventing client inference attacks | Publicación