Tag
This paper demonstrates that formal robustness certificates for embedded neural interface models can pass even when task accuracy collapses under adversarial attack, and proposes a unified empirical audit framework to address alignment failures between training objectives and operational user welfare.
This paper introduces Synergistic Simplex, a new runtime assurance architecture for autonomous systems that allows safety monitors to use ML outputs while preserving formal safety guarantees. The authors demonstrate its effectiveness in improving performance for obstacle detection in autonomous vehicles.