The professorship of Reinforcement Learning and Computational Decision-Making deals with the research and development of novel methodologies for obtaining effective strategies to solve Decision-Making problems and control tasks in simulation and in the real world. We strive for making Reinforcement Learning more efficient and practical, in order to fully express its potential on challenging and impactful real-world problems. In particular, we investigate lightweight methods by addressing open questions in transfer, curriculum, and multi-task learning, where we aim at obtaining strategies that generalise across different problems and seamlessly adapt to previously unseen ones. We conduct our research under a both theoretical and empirical lens, where the rigorous theoretical treatment of the proposed methodologies is complemented with empirical evidence of their effectiveness on simulated benchmarks and real-world applications, e.g., finance, games, robotics.
Our Chair adopts a strongly interdisciplinary approach to face the heterogeneous challenges arising from the application of Reinforcement Learning in disparate fields of the real world. At the same time, we conduct our investigation under a rigorous theoretical lens inspired by established results in classical Reinforcement Learning literature. Our research contribution and methodological advances have been published in leading machine learning and robotics conferences (NeurIPS, ICML, ICLR, AAAI, RSS, ICRA, ...) and journals (JMLR, Frontiers in Robotics and AI).