Αίθουσα Β1.101 (αίθουσα τηλεκπαίδευσης ΗΜΜΥ), 11:00πμ
Τίτλος: Efficient Reinforcement Learning
Ομιλητής: Χρήστος Δημητρακάκης, University of Amsterdam
Recent advances in reinforcement learning methods have made them more applicable to sequential decision making tasks. This talk gives a brief introduction to efficient reinforcement learning and outlines my current work in the area: learning in continuous spaces, focusing on approximate policy iteration and tree search.
C.Dimitrakakis has graduated from the University of Manchester and obtained his MSc from the University of Essex in 1998. After a brief break for military service and work in industry, he started a PhD in IDIAP/EPFL. His thesis, entitled "Ensembles for Sequence Learning", investigated the use of ensemble methods in speech recognition and reinforcement learning. His first post-doc was at the University of Leoben, where he worked on continuous bandit problems and active learning. At the same time, he started a collaboration with Dr Lagoudakis at TUC on approximate policy iteration. He is currently a post-doc at the University of Amsterdam, continuing work on online reinforcement learning. On and off, he has been working on the simulation and AI-drivers for the open source car racing simulator TORCS where he found that the simplest things often work best.