A High-Level Language for Neural Network Verification… and more!

In this talk I will discuss Vehicle, an advanced domain-specific language designed to embed logical specifications into neural networks, enabling robust verification and training processes. By providing a unified interface, Vehicle streamlines complex tasks such as neural network verification, property-driven training, and integration with proof assistants. Moreover, Vehicle supports diverse applications, from classification robustness in image recognition to ensuring stability in automated control systems, such as autonomous vehicle navigation under variable environmental conditions. Vehicle's compatibility with multi-backend architectures further promotes versatility, making it a powerful tool for developers and researchers aiming to create verifiable, reliable, and adaptable AI systems. The talk will discuss various application domains as well as recent advances.