Beyond Weights and Gradients: A Taxonomy of Federated Learning Messages
A new paper proposes a formal definition and taxonomy for federated learning messages, moving beyond traditional model weights and gradients. The research categorizes these exchanges into model structures, statistical summaries, and data-conditioned representations, analyzing their computational demands, communication costs, and privacy risks. The authors note a significant shift in recent publications towards more diverse messaging paradigms in federated learning since 2021. AI
IMPACT Provides a structured framework for optimizing federated systems and understanding trade-offs in decentralized training.