2025 | CoVernet

Machine learning is increasingly deployed in safety-critical applications; however, its verification and certification remain major challenge. Traditional testing methods, like neuron coverage, do not show whether a neural network has been fully tested across its operational design domain (ODD).

CoVerNet introduces a new framework for verifying and testing neural networks used in safety-critical areas like aviation. CoVerNet measures model reliability using a new idea called behavioral coverage, which checks how thoroughly a network’s decision space has been explored.

CoVerNet aims to create a full framework for verifying and validating neural networks used in safety-critical aerospace systems. The project introduces the concept of behavioral coverage, which measures how thoroughly a network's decision-making behavior has been tested. CoVerNet's ultimate goal is to ensure that machine learning models meet the needs for traceability, robustness, and safety required for aviation certification.

The project will bring enhanced support for certification processes, bridging the gap between AI testing and safety standards such as DO-178C. It can be seen as a step toward trustworthy and certifiable AI, paving the way for safer deployment of neural-network-based systems in aviation and beyond.