Utility Optimization of Federated Learning with Differential Privacy

Table 1

Parameters setting on different datasets.

Parameters	MNIST	CIFAR-10
	1.0	1.0
	0.7	0.7
Batch size	20	20
Clipping threshold	4.0	7.0
Learning rate	0.05	0.05
Number of participants	1000	100
Training round	50	50