Mahyar Fazlyab

Mathematical Institute of the Slovak Academy of Sciences, Johns Hopkins University, University of Pennsylvania

Papers

Total Citations

H-Index

About

Mahyar Fazlyab is a leading researcher at the intersection of control theory, optimization, and machine learning, with a primary focus on ensuring the safety and stability of autonomous systems. His work is central to the emerging field of safe AI, where he develops rigorous mathematical frameworks to verify and guarantee the behavior of neural network-controlled dynamical systems. A key contribution is his development of "one-shot" reachability analysis, a method that efficiently computes the set of all possible states a neural network dynamical system can reach, enabling formal safety verification without costly iterative simulations. This work, alongside his research on stability analysis for systems with neural network controllers and complementarity problems (crucial for robotics contact dynamics), has garnered significant attention. Fazlyab has also pioneered "safe physics-informed machine learning," a paradigm that integrates physical laws into learning-based controllers to provide formal safety guarantees. His earlier work on prediction-correction interior-point methods for time-varying convex optimization laid the theoretical groundwork for tracking optimal solutions in dynamic environments. With over 20 citations across his most prominent works, Fazlyab is shaping the future of safe and reliable autonomy.