Communication-efficient Federated Learning with Privacy Enhancing via Probabilistic Scheduling

Federated Learning (FL) has been recognized as a key technology for enabling Artificial Intelligence (AI) services in 5G networks due to its distributed nature that effectively addresses privacy concerns and reduces transmission costs. However, the performance of wireless FL systems is often affected by communication delays and artificial noise added to protect differential privacy (DP). In this paper, we propose a design methodology for joint device access probability and artificial Gaussian noise that strikes a balance between training time and privacy protection. We describe the convergence behavior and DP amplification properties of FL, and then achieve the optimal device access probability and ensure the appropriate artificial Gaussian noise by minimizing the training time under DP constraints. In addition, numerical results validate the theoretical analysis and demonstrate the effectiveness of the proposed method.