Adding perturbations directly to every hidden embedding of a GNN is claimed to subsume existing feature-, edge-, and weight-perturbation defenses, but the claim rests on simplifications that the experiments do not actually test.
Title resolution pending
1 Pith paper cite this work. Polarity classification is still indexing.
1
Pith paper citing it
fields
cs.LG 1years
2025 1verdicts
REJECT 1representative citing papers
citing papers explorer
-
Unifying Adversarial Perturbation for Graph Neural Networks
Adding perturbations directly to every hidden embedding of a GNN is claimed to subsume existing feature-, edge-, and weight-perturbation defenses, but the claim rests on simplifications that the experiments do not actually test.