3–6 month project advancing federated learning for on-device model training: privacy-preserving, communication-efficient algorithms and scalable distributed optimization for mobile and edge systems, suited for students and early-career researchers in ML and related fields.