Verifying Graph Neural Networks with Readout is Intractable
2510.08045v1
cs.LO, cs.AI, cs.CC, cs.LG
2025-10-11
Авторы:
Artem Chernobrovkin, Marco Sälzer, François Schwarzentruber, Nicolas Troquard
Abstract
We introduce a logical language for reasoning about quantized
aggregate-combine graph neural networks with global readout (ACR-GNNs). We
provide a logical characterization and use it to prove that verification tasks
for quantized GNNs with readout are (co)NEXPTIME-complete. This result implies
that the verification of quantized GNNs is computationally intractable,
prompting substantial research efforts toward ensuring the safety of GNN-based
systems. We also experimentally demonstrate that quantized ACR-GNN models are
lightweight while maintaining good accuracy and generalization capabilities
with respect to non-quantized models.