Assess students' ability to design a formally verifiable, Byzantine-resilient federated learning protocol that provides differential privacy guarantees and certified robustness for training deep neural networks on heterogeneous, non‑i.i.d. edge devices. The task should require: precise system and threat models; protocol specification (peer/client roles, secure aggregation, robust aggregation rules); a privacy mechanism and accounting method (e.g., DP-SGD, Rényi/moments accountant) with formal privacy proofs; certified-robust training or certification technique (e.g., randomized smoothing, certified defenses) with guarantees under Byzantine behaviors; formal verification artifacts or proof sketches (theorem prover, model checker, or SMT); convergence/utility analysis under heterogeneity; an experimental evaluation plan (datasets, metrics, Byzantine scenarios, privacy-utility-robustness trade-offs); and considerations for performance, scalability, and real-world deployment.