Abstract

Future cyber-physical systems will employ complex learning-based sensing, computation, and communication components for their perception, planning, control, and coordination, and could operate in highly dynamic and uncertain environment with safety and security assurance. To realize this vision, we have to better understand and address the challenges from the “unknowns” – the inherent uncertainties from system inputs and model inaccuracies, the unexpected disturbances from component failures, environment interferences, and malicious attacks, and the lack of analyzability of neural network-based machine learning techniques. In this talk, I will discuss these challenges and present some of our recent work to address these challenges in a holistic manner, including 1) end-to-end verification, design and adaptation methods for ensuring robust and safe application of neural networks in perception and decision making; and 2) cross-layer methods for mitigating execution disturbances (e.g., timing violations, soft errors, malicious attacks) using weakly-hard paradigm.