Table 1.

Performance (%) comparison under the heterogeneous cross-device settings.

Agg. Method	Model	FMNIST			CIFAR-10			CIFAR-100
		$\alpha =1$	$\alpha =0.5$	$\alpha =0.1$	$\alpha =1$	$\alpha =0.5$	$\alpha =0.1$	$\alpha =1$	$\alpha =0.5$	$\alpha =0.1$
✘	FedProto	85.05	87.66	89.04	76.59	78.17	82.96	58.03	66.31	68.60
✘	FML	86.54	90.71	92.63	80.80	85.24	88.58	58.77	66.90	68.74
FedAvg	FedType global	84.11	83.93	81.32	66.40	63.39	58.17	38.36	38.17	35.45
	FedType proxy	86.09	89.45	93.16	80.65	82.57	85.04	56.24	61.06	62.31
	FedType private	87.26	91.22	94.77	82.56	86.83	91.90	57.33	65.69	68.14
FedProx	FedType global	86.96	86.44	84.29	68.26	65.86	63.75	41.88	39.31	36.53
	FedType proxy	87.03	91.50	92.64	82.19	82.48	87.80	58.56	61.22	62.64
	FedType private	87.65	93.84	94.98	83.69	86.92	92.03	59.18	65.45	68.37
pFedMe	FedType global	87.82	87.13	85.86	68.71	65.22	64.95	41.55	40.92	38.60
	FedType proxy	88.63	92.05	93.38	82.64	83.00	88.14	59.04	62.68	64.89
	FedType private	88.96	92.36	94.86	83.47	87.24	92.16	59.78	67.07	69.51
pFedBayes	FedType global	88.20	87.85	86.04	68.41	66.87	63.32	43.73	41.24	38.72
	FedType proxy	89.69	92.11	93.29	83.33	84.49	89.10	59.47	62.96	63.51
	FedType private	90.26	93.17	95.88	84.09	88.67	92.38	59.62	67.35	69.60