Using Decentralized Aggregation for Federated Learning with Differential Privacy