Verifying Quantized GNNs With Readout Is Decidable But Highly Intractable

Fuente: arXiv
Gespeichert in:
Bibliographische Detailangaben
Hauptverfasser: Chernobrovkin, Artem, Sälzer, Marco, Schwarzentruber, François, Troquard, Nicolas
Format: Preprint
Veröffentlicht: 2025
Schlagworte:
Online-Zugang:
Tags: Tag hinzufügen
Keine Tags, Fügen Sie den ersten Tag hinzu!
_version_ 1866917435604467712
author Chernobrovkin, Artem
Sälzer, Marco
Schwarzentruber, François
Troquard, Nicolas
author_facet Chernobrovkin, Artem
Sälzer, Marco
Schwarzentruber, François
Troquard, Nicolas
contents 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.
format Preprint
id arxiv_https___arxiv_org_abs_2510_08045
institution arXiv
publishDate 2025
record_format arxiv
spellingShingle Verifying Quantized GNNs With Readout Is Decidable But Highly Intractable
Chernobrovkin, Artem
Sälzer, Marco
Schwarzentruber, François
Troquard, Nicolas
Logic in Computer Science
Artificial Intelligence
Computational Complexity
Machine Learning
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.
title Verifying Quantized GNNs With Readout Is Decidable But Highly Intractable
topic Logic in Computer Science
Artificial Intelligence
Computational Complexity
Machine Learning
url https://arxiv.org/abs/2510.08045