An Adaptive Communication-Efficient Federated Learning to Resist Gradient-Based Reconstruction Attacks

Table 2

The evaluation metrics of the DLG attacks with the CIFAR100 dataset in different frequencies.

Method	Iterations	Communication frequency	SSIM	MSE
DLG	500	1	0.9985	0.3300
DLG	500	2	0.9961	0.5437
DLG	500	3	0.9892	1.4399
DLG	500	4	0.9815	2.7646
DLG	500	5	0.9731	4.3394
DLG	500	6	0.9631	6.3177
DLG	500	7	0.9521	8.3981
DLG	500	8	0.9427	10.8416
DLG	500	9	0.9325	12.7527
DLG	500	10	0.9215	15.2681
DLG	500	15	0.8824	23.8498
DLG	500	20	0.8427	30.3106
DLG	500	30	0.7856	36.5652
DLG	500	40	0.7440	40.5796
DLG	500	50	0.7170	42.4485