Assess the learner's ability to design a graduate-level federated learning system for heterogeneous edge devices that is privacy-preserving, communication- and compute-efficient, and personalized. The task should require specification and justification of chosen secure aggregation and differential privacy mechanisms (including noise calibration and privacy budget accounting), personalized model distillation strategies for on-device personalization, approaches to handle device heterogeneity (compute, communication, and data distribution), compression and communication-reduction techniques, and defenses against adversarial or drop-out behaviors. Expect analytical trade-off discussion (utility vs. privacy vs. cost), a protocol-level design (algorithms, crypto primitives, and interaction rounds), complexity and convergence analysis, and a concrete evaluation plan (datasets, metrics, baselines, simulation or deployment considerations).