Broadly speaking, I am interested in the algorithmic and computational aspects of Artificial Intelligence, with a particular emphasis on scalable and efficient approaches for online Reinforcement Learning agents.
Consider an agent let loose in an unknown world with only a crude reward signal and limited sensor information as feedback. An agent that can be expected to accumulate a lot of reward in many different worlds can be viewed as being generally intelligent. How might we go about designing such an agent?
Universal Artificial Intelligence
Game Tree Search
Machine Learning for Games