Wednesday, February 4, 2026 | 9:30 - 11:00 AM
Department of Electronics, Information and Bioengineering - Politecnico di Milano
BIO1 Room (Building 21) | Online on Teams
Speaker: Prof. Ezio Bartocci (TU Wien)
Department of Electronics, Information and Bioengineering - Politecnico di Milano
BIO1 Room (Building 21) | Online on Teams
Speaker: Prof. Ezio Bartocci (TU Wien)
Contacts: Bruno Guindani | bruno.guindani@polimi.it
Abstract
Artificial intelligence is rapidly becoming a core component of cyber-physical systems, enabling autonomy and adaptability in domains such as robotics, autonomous driving, and intelligent edge platforms. However, the use of machine learning and reinforcement learning in safety-critical settings raises fundamental challenges related to reliability, robustness, and uncertainty.
In this seminar, Prof. Ezio Bartocci will present his recent research in formal methods for AI-enabled cyber-physical systems, focusing on the verification, monitoring, and testing of learning-based components. The talk will cover formal reasoning about neural networks under uncertainty, quantitative evaluation of learning-enabled behaviors against temporal specifications, and runtime monitoring and assurance mechanisms for AI-driven systems.
He will further discuss how formal constraints and rules can be integrated into reinforcement learning, enabling safer and more interpretable policy evaluation and improvement. Complementing these approaches, he will present automated techniques for probabilistic program analysis that provide exact or bounded guarantees on distributions and moments without relying on sampling. The seminar concludes by highlighting how these methods support the deployment of robust, trustworthy, and resource-aware AI in real-world cyber-physical systems, and by outlining open research challenges at the intersection of machine learning, control, and formal verification.