notesum.ai

Published at December 10

AHSG: Adversarial Attacks on High-level Semantics in Graph Neural Networks

cs.LG

Released Date: December 10, 2024

Authors: Kai Yuan¹, Xiaobing Pei¹, Haoran Yang¹

Aff.: ¹Huazhong University of Science and Technology

Arxiv: http://arxiv.org/pdf/2412.07468v1

Refer to caption

Dataset	Method	Clean	Random	DICE	Meta-Self	GradArgmax	PGD	AHSG
Cora	GCN	0.8230	0.8090	0.7970	0.8090	0.7160	0.7080	0.6460
	Jaccard	0.7860	0.7840	0.7630	0.7360	0.7120	0.7590	0.6760
	Svd	0.7290	0.7110	0.7070	0.6650	0.6730	0.6860	0.6480
	ProGNN	0.8090	0.7940	0.7740	0.6590	0.7370	0.7030	0.6540
	SimPGCN	0.7900	0.7820	0.7770	0.7840	0.7030	0.7280	0.6820
	RGCN	0.8010	0.7480	0.7450	0.7530	0.6900	0.6780	0.6510
Citeseer	GCN	0.6660	0.6550	0.6470	0.6500	0.5780	0.5930	0.5280
	Jaccard	0.6650	0.6320	0.6490	0.6450	0.6170	0.6230	0.5760
	Svd	0.6010	0.5930	0.5740	0.5730	0.5880	0.5650	0.5580
	ProGNN	0.6830	0.6250	0.6250	0.5310	0.5660	0.6030	0.5300
	SimPGCN	0.6560	0.6460	0.6500	0.6480	0.5850	0.6150	0.5810
	RGCN	0.6100	0.5790	0.5800	0.5690	0.5760	0.5730	0.5340
Cora-ML	GCN	0.8590	0.8510	0.8490	0.8270	0.7430	0.7910	0.7330
	Jaccard	0.8600	0.8390	0.8360	0.8070	0.7700	0.8280	0.7620
	Svd	0.8200	0.7990	0.8260	0.7940	0.7590	0.8160	0.7570
	ProGNN	0.8400	0.8050	0.8050	0.7910	0.7490	0.8230	0.7350
	SimPGCN	0.8460	0.8360	0.8390	0.8170	0.7430	0.8180	0.7400
	RGCN	0.8590	0.8420	0.8420	0.8310	0.7540	0.8280	0.7550