Reinforcement Learning · CAIDAS · University of Würzburg
Teaching machines to learn from interaction.
LiteRL — the Reinforcement Learning and Computational Decision-Making group — develops methods for finding effective strategies in decision-making and control, to make reinforcement learning more efficient and practical. Led by Prof. Carlo D'Eramo.
About the lab
What we work on
We study how autonomous agents can efficiently acquire expert skills while accounting for the complexity of the real world. Our focus is on lightweight methods for transfer, curriculum and multi-task learning, combining theoretical analysis with empirical evaluation, with applications in finance, games and robotics. Our work appears at venues such as NeurIPS, ICML, ICLR, AAAI, RSS and ICRA, and in JMLR.
Multi-task & transfer learning
Agents that share knowledge across tasks instead of learning each one from scratch.
Curriculum learning
Ordering training experience so that agents build skills progressively.
Multi-agent, adversarial & options
Learning with other agents, against opponents, and with reusable temporally extended skills.
Want to work with us?
We welcome Ph.D. candidates, postdocs, research assistants and thesis students who want to make reinforcement learning efficient and practical.