Toolbox for Verification of Autonomous Systems with Neural Components (TOOL-VAN)

As the Navy seeks to deploy neural networks for perception and control capabilities in autonomous systems, they must ensure networks behave predictably and reliably. Researchers have responded to this need by producing a proliferation of methods for formally verifying properties of neural networks. However, applying these methods is a challenge, as methods vary in their applicability to different architectures, properties, and domains, requiring a prohibitively high level of technical expertise for successful use. To address the need for user-friendly verification methods, Charles River Analytics proposes to design and demonstrate the feasibility of a Toolbox for Verification of Autonomous Systems with Neural Components (TOOL-VAN). TOOL-VAN is a general-purpose toolbox using automated verification method selection, specification-based retraining, cyber¢physical system integration, and an intuitive graphical user interface to allow users of different backgrounds to verify autonomous systems with neural components in safety, mission, and time-critical scenarios. TOOL-VAN will lower the barrier of entry to neural network verification, enabling the safe deployment of artificial intelligence (AI) systems in the Navy and throughout the Federal Government.