Deep Reinforcement Learning for Human-Level Agents

Deep learning has enabled traditional reinforcement learning methods to deal with high-dimensional problems. However, there have been several issues with deep reinforcement learning methods such as the limited exploration capability of learning agents, the ability to integrate human knowledge into learning agents or their extendibility to handle multi-objective problems as well as multi-agent control. In this talk, I will discuss about these issues and present my recent studies on deep reinforcement learning to address these issues.