Privacy-Preserving Federated Graph Neural Network Learning on Non-IID Graph Data

Table 1

The results of the federated DeepWalk framework. LOC indicates the result of local training, FED indicates the result of federated learning, and GLOBAL indicates the cumulative improvement of all clients.

Dataset (classifier)	Client 1		Client 2		Client 3		Client 4		GLOBAL
	LOC	FED	LOC	FED	LOC	FED	LOC	FED
Cora (MLP)	79.1	80.1	76.0	76.0	74.3	76.1	75.0	76.6	+4.41
Cora (SVC)	79.3	81.4	77.5	78.7	75.5	77.9	75.0	78.0	+8.72
CiteSeer (MLP)	48.1	51.4	46.4	50.4	48.9	51.6	49.2	51.7	+12.5
CiteSeer (SVC)	56.3	60.6	53.5	58.6	57.3	61.6	55.3	60.4	+18.8
Cora (full)	MLP		79.9		SVC		82.0
CiteSeer (full)	MLP		58.6		SVC		65.4