Assess students' ability to evaluate privacy–utility trade-offs in federated learning when differential privacy is applied and adversaries are realistic. Tasks include critiquing DP mechanisms (local vs. central DP, noise calibration, clipping), quantifying impacts on accuracy and convergence, designing experiments and metrics to measure privacy leakage (membership/attribute inference) and robustness (model poisoning, backdoors), analyzing attacker capabilities and threat models, and recommending parameter/settings or mitigation strategies that balance utility and risk given specified threat assumptions.