Title: Deep reinforcement learning for controlling complex systems, by Alexander Kuhnle, University of Cambridge
Abstract: Reinforcement learning is concerned with optimal decision making in a complex environment, and has recently been successfully combined with deep learning to beat humans at the board game Go. Inspired by this landmark accomplishment, researchers and practitioners started applying these methods to a range of real-world industrial control problems like, for instance, in robotics. In this talk, I will give a high-level overview of deep reinforcement learning, discuss both motivation and problems of applying such techniques to system control, and present promising recent approaches to overcome these hurdles.