Encoded Gradients Aggregation against Gradient Leakage in Federated Learning