- This event has passed.
Csaba SZEPESVARI (University of Alberta) – "Towards Efficient Planning in Large Markov Decision Processes"
Time: 4:00 pm – 5:15 pm
Date: 29th of March 2021
Csaba SZEPESVARI (University of Alberta) – “Towards Efficient Planning in Large Markov Decision Processes”
Abstract: During the last decade reinforcement learning algorithms achieved outcomes far beyond expectations and not only in one area, but in a diverse set of areas, such as control, robotics, electronics, or computer games, just to list a few. What is common in the algorithms powering these breakthroughs that they all use powerful computers, simulators and reinforcement learning algorithms with one common key component: function approximation. Formally, these algorithms solve planning problems in large Markov decision processes where intractabality is avoided by using function approximation. Why, when and how these algorithms succeed is the central questions of this talk, in which I report on recent theoretical advances made on large scale planning in the presence of function approximation. The emerging picture is intriguingly complex: seemingly small changes to the way a problem is posed can cause a provably tractable problem to become intractable and vice versa. In this talk I will summarize the most striking of these results and will discuss a few of fascinating open questions.
Cristina BUTUCEA (CREST), Alexandre TSYBAKOV (CREST), Karim LOUNICI (CMAP) , Zoltan SZABO (CMAP)