Assess students' ability to apply differential privacy (DP) mechanisms to federated learning update aggregation: select and justify clipping norms, calibrate Gaussian/Laplace noise for client- or record-level DP, use privacy accounting methods (e.g., moments accountant, RDP) to compute (ε,δ) under composition, evaluate effects on model utility and convergence, and describe practical secure-aggregation and communication trade-offs in experiments or algorithm design.