Talking with Verifiers: Automatic Specification Generation for Neural Network Verification

Abstract

Neural network verification tools currently support only a narrow class of specifications, typically expressed as low-level constraints over raw inputs and outputs. This limitation significantly hinders their adoption and practical applicability across diverse application domains where correctness requirements are naturally expressed at a higher semantic level. This challenge is rooted in the inherent nature of deep neural networks, which learn internal representations that lack an explicit mapping to human-understandable features. We begin addressing this gap by introducing a novel component to the verification pipeline, making existing verification tools applicable to a broader range of domains and specification styles. Our framework enables users to formulate specifications in natural language, which are then automatically analyzed and translated into formal verification queries compatible with state-of-the-art neural network verifiers. The verifier’s formal guarantee applies to the generated low-level specification after the parsed intent and grounding have been approved. We evaluate our approach on both structured (tabular) and unstructured (image) datasets, demonstrating that it successfully verifies complex semantic specifications that were previously much less accessible to laymen. Our results show that this translation process maintains high fidelity to user intent while incurring low computational overhead, thereby substantially extending the applicability of formal neural network verification to real-world, high-level requirements.