Figure 1: the FedAvg algorithm. The aggregator distributes the current global model to clients, the clients perform several iterations of training, each on their own local data, and then the aggregator averages the resulting model updates to form the next