ABSTRACT:

 Model based reinforcement learning refers to reinforcement learning methods that explicitly construct and reason with models of the environment of the learning agent. There are many reasons to believe that model based reinforcement learning is crucial for increasing the flexibility and data efficiency of reinforcement learning methods. In particular, models can be used in a planning or inference process, which can increase the range of policies that an agent can represent, while models can also retain crucial aspects of past experiences. Equally importantly, principled approaches to exploration (optimism) is arguably a more natural fit to model based RL than to model-free RL. Yet, model based RL methods rarely make to the very top of leaderboards. In this talk, I will review the reasons behind why is it challenging to construct efficient model based RL methods and describe some novel developments which may hold the promise to change the present poor record of model based RL methods.

BIO:

 Csaba Szepesvári is a Canada CIFAR AI Chair, the team-lead for the “Foundations” team at DeepMind and a Professor of Computing Science at the University of Alberta. He earned his PhD in 1999 from Jozsef Attila University, in Szeged, Hungary. He has authored three books and about 200 peer-reviewed journal and conference papers. He serves as the action editor of the Journal of Machine Learning Research and Machine Learning, as well as on various program committees. Dr. Szepesvari’s interest is artificial intelligence (AI) and, in particular, principled approaches to AI that use machine learning. He is the co-inventor of UCT, a widely successful Monte-Carlo tree search algorithm. UCT ignited much work in AI, such as DeepMind’s AlphaGo which defeated the top Go professional Lee Sedol in a landmark game. This work on UCT won the 2016 test-of-time award at ECML/PKDD.

Hosted by Professor Quanquan Gu