Hanna Krasowski
University of California, Berkeley
krasowski@berkeley.edu
Bio
Hanna Krasowski is a postdoc in the EECS department at the University of California, Berkeley, and affiliated with the Berkeley Artificial Intelligence Research Lab. Her research combines formal methods and machine learning to develop safe and data-efficient models for real-world systems, with applications in maritime motion planning and biomolecular modeling. She earned her Ph.D. from the Technical University of Munich in 2024, working in the Cyber-Physical Systems group, and was a visiting researcher in the AMBER lab at Caltech in 2022. She is the principal developer of CommonOcean, a benchmarking and software suite for maritime navigation.
Areas of Research
- Artificial Intelligence
Specification Guards for Neuro-Symbolic Autonomy in Data-Limited and Dynamic Environments
While machine learning has made rapid progress in virtual environments, its influence on real-world cyber-physical systems is still narrow in scope, especially in dynamic and uncertain settings. I believe real-world autonomy requires combining symbolic and learning-based methods to leverage abstract knowledge and make informed and safe decisions from limited data. My research advances this vision by developing specification guards for safe, interpretable, and efficient learning. The first thrust of my research develops algorithms for safe and interpretable machine learning. I focus on provably safe reinforcement learning algorithms that provide hard guarantees for complex specifications during training and deployment in continuous domain systems. My second thrust uses formal methods, i.e., set-based reachability analysis and temporal logic, to make abstract domain knowledge computationally tractable and guide machine learning algorithms. I demonstrate real-world readiness on various applications, including autonomous vehicles, biomolecular processes, and maritime navigation. For maritime navigation, I introduced CommonOcean, a dedicated software and benchmarking platform for maritime vessel navigation. Ultimately, I envision that advancing machine learning with formal methods enables the development of a foundational framework for real‑world autonomy capable to tap unstructured and multi-modal information sources, including implicit and mathematical representations of system models as well as time-series, text, or traffic data.