AI Thrust Seminar | Learning-Augmented Algorithms for Cyber-Physical Systems: Limitations, Solutions, and Future Directions

The integration of AI techniques into real-world physical systems has garnered significant interest in recent years, with the promise of providing new data-driven solutions to real-world problems. However, this integration also poses practical issues such as reliability, scalability, and trustworthiness for AI-integrated algorithms. In this talk, we will present a framework that strikes a balance between the robustness and consistency tradeoff when designing online algorithms with machine learning predictions, and discuss its limitations. Using discrete control models, we will demonstrate potential solutions and their applications in smart grids. Finally, we will explore intriguing research directions for future work in this rapidly evolving field. The goal of this introductory talk is to shed light on the challenges and opportunities of AI integration into real-world cyber-physical systems and inspire further research in this area.