Bridging Control with Neural Network Verifier alpha-beta-CROWN: A Tutorial
A tutorial presents a unified framework that bridges control theory with the neural network verifier alpha-beta-CROWN. The framework aims to formally verify properties like stability and safety in learning-based controllers, which are increasingly used in safety-critical domains such as autonomous driving, robotics, and power systems. Existing verification approaches often rely on specific structural assumptions or scale poorly with high-dimensional neural networks. Alpha-beta-CROWN serves as a general-purpose bounding engine for nonlinear functions, enabling more scalable and transferable verification. The tutorial is published on arXiv with ID 2605.26577.
Key facts
- Learning-based controllers are popular for their expressiveness and empirical performance.
- Formal verification of controller properties is needed for safety-critical scenarios.
- Prior verification approaches are tied to specific structural assumptions or scale poorly.
- The tutorial presents a unified framework bridging control with alpha-beta-CROWN.
- Alpha-beta-CROWN is a general-purpose bounding engine for nonlinear functions.
- The framework aims to improve scalability and transferability of verification.
- The tutorial is available on arXiv with ID 2605.26577.
- Safety-critical domains include autonomous driving, robotics, and power systems.
Entities
Institutions
- arXiv