Deep Reinforcement Learning
Reinforcement learning is the science of how to learn to make decisions through interaction. Deep learning is the science of learning representations, mappings, and functions from data. The intersection of these two fields has great potential, because deep reinforcement learning algorithms give us a means to learn behaviour through interaction with complex problems, by using general learning algorithms rather than having to rely on extensive domain knowledge. In my talk, I will discuss why I'm excited about this research field, I will talk about open challenges, and discuss some recent successes in applying these algorithms to interesting problems.
Hado van Hasselt is a Senior Staff Research Scientist at DeepMind. He joined DeepMind in 2014, after doing a postdoc at the University of Alberta with Rich Sutton. He obtained his Ph.D. in artificial intelligence with a dissertation on reinforcement learning at Utrecht University in the Netherlands. His research focuses mostly on core learning algorithms that have the potential to scale to rich and interesting domains, and on the intersection of reinforcement learning and deep learning.