Benchmarking Graph Neural Networks in Solving Hard Constraint Satisfaction Problems
Disordered Systems and Neural Networks
2026-03-12 v2 Machine Learning
Abstract
Graph neural networks (GNNs) are increasingly applied to hard optimization problems, often claiming superiority over classical heuristics. However, such claims risk being unsolid due to a lack of standard benchmarks on truly hard instances. From a statistical physics perspective, we propose new hard benchmarks based on random problems. We provide these benchmarks, along with performance results from both classical heuristics and GNNs. Our fair comparison shows that classical algorithms still outperform GNNs. We discuss the challenges for neural networks in this domain. Future claims of superiority can be made more robust using our benchmarks, available at https://github.com/ArtLabBocconi/RandCSPBench.
Cite
@article{arxiv.2602.18419,
title = {Benchmarking Graph Neural Networks in Solving Hard Constraint Satisfaction Problems},
author = {Geri Skenderi and Lorenzo Buffoni and Francesco D'Amico and David Machado and Raffaele Marino and Matteo Negri and Federico Ricci-Tersenghi and Carlo Lucibello and Maria Chiara Angelini},
journal= {arXiv preprint arXiv:2602.18419},
year = {2026}
}