[C1] Communication-Efficient Learning of Deep Networks from Decentralized Data AISTATS 17
This is the first paper to give name to federated learning. It explores the benefits of applying additional computation to reduce communication as FedAvg, it also do good to unbalanced data.