A New Strategy for Verifying Reach-Avoid Specifications in Neural Feedback Systems
This work addresses verification challenges for neural feedback systems, which is incremental as it builds on existing forward analysis methods.
The paper tackled the limited scalability of backward reachability methods for verifying reach-avoid properties in neural feedback systems by introducing new algorithms for over- and under-approximations and integrating them with forward analysis into a unified framework.
Forward reachability analysis is the predominant approach for verifying reach-avoid properties in neural feedback systems (dynamical systems controlled by neural networks). This dominance stems from the limited scalability of existing backward reachability methods. In this work, we introduce new algorithms that compute both over- and under-approximations of backward reachable sets for such systems. We further integrate these backward algorithms with established forward analysis techniques to yield a unified verification framework for neural feedback systems.