TRAILS Trustworthy Machine Learning Seminar Series
Speaker: Calin Belta, Brendan Iribe Endowed Professor of Electrical and Computer Engineering and Computer Science at UMD
Where: University of Maryland, CSIC 2120 and on Zoom: https://umd.zoom.us/j/93960331973?pwd=tVQ7gtFucJEkSYna38zePzkPpHm3J7.1
Talk: Formal Methods for Safe and Interpretable Reinforcement Learning
Abstract: Modern engineering applications such as autonomous robots involve large-scale dynamical systems whose models are complex, uncertain, or only partially known, placing them beyond the reach of traditional model-based control techniques. Data-driven approaches, including machine learning and reinforcement learning, offer powerful tools for learning and decision-making in these settings, but often lack interpretability and provide limited guarantees of correctness and safety—critical requirements. Formal methods address these shortcomings by providing rich languages for expressing complex objectives and safety requirements, algorithmic techniques for verification and synthesis, and proofs that the resulting systems behave as intended. This talk that brings together control theory, formal methods, and machine learning to create control systems that scale and adapt to uncertain or unknown dynamics while remaining safe, correct, and interpretable. Examples from autonomous driving and robotic manipulation illustrate how this integration enables the synthesis of controllers for complex, safety-critical tasks.
[1] A formal methods approach to interpretable reinforcement learning for robotic planning
[2] Safe Model-based Control from Signal Temporal Logic Specifications Using Recurrent Neural Networks
Bio: Calin Belta is the Brendan Iribe Endowed Professor of Electrical and Computer Engineering and Computer Science at the University of Maryland, College Park, where is also part of the Maryland Robotics Center (MRC) and the Institute for Systems Research (ISR). His research focuses on making control and machine learning systems safe and Interpretable, with particular emphasis on robotics and systems biology. Notable awards include the AFOSR YIP, the NSF CAREER award, and the HSCC Test of Time award. He is a Fellow of the IEEE.
About the TRAILS Trustworthy Machine Learning Seminar Series: Each week a researcher from across the TRAILS community presents current work on what it takes to make AI systems worth trusting: measuring and evaluating their behavior, interpreting and steering them, keeping them robust and secure, protecting privacy, and understanding how people actually rely on them — with occasional excursions into the philosophy and psychology of trust. Talks are 40 minutes, followed by 20 minutes of discussion.