TNODEV: Toolbox for Neural ODE Verification

A research team has released TNODEV, described as the first sound formal verifier for neural ordinary differential equations, a class of machine-learning models increasingly deployed in safety-critical systems such as continuous-time controllers and automated decision pipelines [1][2]. The tool, detailed in a paper submitted to the arXiv preprint repository on June 15, 2026, integrates a falsification checker, a fast interval-based reachability backend, and a verification-and-refinement loop with three input-set splitting heuristics into a single pipeline [1][2]. TNODEV supports safe-set inclusion verification for pure neural ODE, neural ODE in closed loop with a neural-network controller, and general neural ODE, with the safe set specifiable as an interval or as the half-space intersection induced by a target classification label [1][2]. The submission, authored by Abdelrahman Sayed Sayed, weighs 1,582 KB [1]. Existing tools dedicated to neural ODE have provided only a single reachability call without iterative input-set refinement, limiting the precision of their verdicts [2]. TNODEV addresses that gap by looping verification and refinement steps and by adding a parallel scheduler [2]. The authors evaluated the verifier on a range of benchmarks covering safe-set inclusion and classification-robustness properties, including a direct reachability comparison against NNV 2.0 and CORA and a verification comparison against NNV 2.0 on MNIST general neural ODE classifiers [1][2]. The paper appears on arXiv, an open-access repository of electronic preprints that is not peer-reviewed but is moderated before posting [6]. Founded in 1991, arXiv passed the two-million-article milestone by the end of 2021 and was receiving roughly 24,000 submissions per month as of November 2024 [6]. The platform also hosts arXivLabs, a framework that allows community collaborators to develop and share experimental tools directly on article pages, such as bibliographic explorers and recommender systems [4][5]. Neural ODE models represent continuous-depth learning architectures that have drawn attention for their ability to model time-series dynamics and control tasks. Their appearance in safety-critical settings has intensified calls for formal verification methods that can provide guarantees about model behavior, rather than relying solely on empirical testing [2]. The TNODEV release marks an effort to supply such guarantees through a sound, end-to-end verification pipeline [2].