{"id":"arxiv/2605.17153","url":"https://arxiv.org/abs/2605.17153","tags":["cs.LG","cs.LO","math.OC"],"score":0,"title":"Stress-Testing Neural Network Verifiers with Provably Robust Instances","domain":"cs.LG","source":"arxiv","authors":["David Troxell","Yulia Alexandr","Sofia Hunt","Stephanie Lei","Guido Montúfar"],"reviews":0,"subject":"cs.LG","summary":"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…","category":"research","comments":0,"disputed":false,"division":"cs","mentions":0,"discussed":0,"published":"2026-05-16","reactions":{}}