Scalable Verification of Nonlinear Neural Feedback Systems via Branch-and-Bound

The proposed framework improves scalability in verifying nonlinear neural feedback systems by using branch-and-bound on system abstractions. It introduces the \rail interface and \clipper algorithm to jointly refine enclosures and split controller activations, preserving symbolic correlations.