Personalized Federated Learning with Semisupervised Distillation

Table 3

Test performance comparison of different optimizer thresholds for FedSD on MNIST and the same model structures.

Optimizer	Learning rate	= 10	= 0.1
SGD	0.01	93.53	41.29
	0.03	87.46	40.71
	0.05	83.22	40.04
	0.001	91.18	41.03
	0.003	90.46	40.28
	0.005	87.66	40.11
Adam	0.001	81.07	39.41
	0.003	82.46	38.78
	0.005	82.42	38.66
	0.0001	86.36	39.65
	0.0003	86.19	39.03
	0.0005	85.82	38.98