Neural Certificate
The use of neural networks to approximate the Hamilton-Jacobi value function or other safety certificate functions, providing provable safety guarantees in continuous state spaces.
Timeline (last 90 days)
The HJ equation has no analytical solution for general nonlinear systems, and numerical solving faces the curse of dimensionality, so recent research has turned to approximating the HJ value function with neural networks (neural certificates).
This work combines formal verification with deep learning, improving the reliability of neural certificates through enumeration-guided training, and holds significant value for safety-critical domains such as autonomous driving and robotics.
All Facts (2)
The HJ equation has no analytical solution for general nonlinear systems, and numerical solving faces the curse of dimensionality, so recent research has turned to approximating the HJ value function with neural networks (neural certificates).
50%UnverifiedThis work combines formal verification with deep learning, improving the reliability of neural certificates through enumeration-guided training, and holds significant value for safety-critical domains such as autonomous driving and robotics.
50%