The FABRIC Strategy for Verifying Neural Feedback Systems

Researchers introduce FABRIC, a novel verification strategy for neural feedback systems that combines forward and backward reachability analysis. Unlike existing approaches focusing on forward analysis, FABRIC addresses scalability limitations by computing over- and under-approximations of backward reachable sets for nonlinear neural networks. This integration significantly advances certification capabilities for safety-critical systems, outperforming previous state-of-the-art methods on...