Test students' ability to design a scalable, privacy-preserving federated learning system for heterogeneous mobile devices: specify an end-to-end architecture, differential privacy mechanisms (noise calibration and accounting), a secure aggregation protocol, client selection and model-update strategies (synchronous/asynchronous, compression, sparsification), methods to handle device heterogeneity and faults (straggler mitigation, dropouts, checkpointing, partial updates), communication and computation trade-offs, threat model and defenses, and a concrete evaluation plan with metrics for privacy-utility trade-offs, convergence, latency, bandwidth use, and robustness; require trade-off analysis, protocol descriptions, and testing scenarios.