Saved in:
Bibliographic Details
Main Authors: Troxell, David, Alexandr, Yulia, Hunt, Sofia, Lei, Stephanie, Montúfar, Guido
Format: Preprint
Published: 2026
Subjects:
Online Access:https://arxiv.org/abs/2605.17153
Tags: Add Tag
No Tags, Be the first to tag this record!
_version_ 1866911691701223424
author Troxell, David
Alexandr, Yulia
Hunt, Sofia
Lei, Stephanie
Montúfar, Guido
author_facet Troxell, David
Alexandr, Yulia
Hunt, Sofia
Lei, Stephanie
Montúfar, Guido
contents Neural network verifiers aim to provide formal guarantees on model behavior, but existing verification benchmarks are fundamentally limited by their lack of ground-truth labels. As a result, verifier evaluation relies on indirect heuristics, which prevents exact scoring and systematic study of verifier failure modes. We address this gap by introducing a reusable framework for generating verification instances whose ground-truth robustness labels are known a priori through analytic construction. Our framework led to the discovery of multiple numeric tolerance concerns and an implementation bug in popular verifiers, highlighting the need for ground-truth labels. Additionally, to systematically study verifier failure modes, we introduce the verification Difficulty Profile, a collection of estimable quantities capturing distinct sources of instance hardness. Using our framework and these profiles, we evaluate five state-of-the-art verifiers and show that different instances stress distinct aspects of the verification pipeline. We show that these results can aid the future development of verifiers as they provide actionable targets for improving numerical reliability, relaxation quality, and search behavior. Our code is publicly available: https://github.com/dtroxell19/VeriStressGT.git.
format Preprint
id arxiv_https___arxiv_org_abs_2605_17153
institution arXiv
publishDate 2026
record_format arxiv
spellingShingle Stress-Testing Neural Network Verifiers with Provably Robust Instances
Troxell, David
Alexandr, Yulia
Hunt, Sofia
Lei, Stephanie
Montúfar, Guido
Machine Learning
Logic in Computer Science
Optimization and Control
Primary 68Q60, Secondary 68T07, 65G20, 90C11, 90C25, 90C57
D.2.4; F.3.1; I.2.6
Neural network verifiers aim to provide formal guarantees on model behavior, but existing verification benchmarks are fundamentally limited by their lack of ground-truth labels. As a result, verifier evaluation relies on indirect heuristics, which prevents exact scoring and systematic study of verifier failure modes. We address this gap by introducing a reusable framework for generating verification instances whose ground-truth robustness labels are known a priori through analytic construction. Our framework led to the discovery of multiple numeric tolerance concerns and an implementation bug in popular verifiers, highlighting the need for ground-truth labels. Additionally, to systematically study verifier failure modes, we introduce the verification Difficulty Profile, a collection of estimable quantities capturing distinct sources of instance hardness. Using our framework and these profiles, we evaluate five state-of-the-art verifiers and show that different instances stress distinct aspects of the verification pipeline. We show that these results can aid the future development of verifiers as they provide actionable targets for improving numerical reliability, relaxation quality, and search behavior. Our code is publicly available: https://github.com/dtroxell19/VeriStressGT.git.
title Stress-Testing Neural Network Verifiers with Provably Robust Instances
topic Machine Learning
Logic in Computer Science
Optimization and Control
Primary 68Q60, Secondary 68T07, 65G20, 90C11, 90C25, 90C57
D.2.4; F.3.1; I.2.6
url https://arxiv.org/abs/2605.17153