A federated learning algorithm, running on DCP, where all the workers return at the same time, regardless of how much they've trained.