Interview with Richard S. Sutton

Sutton: Well, of course there have been many. In no particular order, the ones that come to mind most quickly are: policy-gradient methods, temporal abstraction methods such as options, and the new predictive ways of thinking about state. Also important are the new Monte Carlo search methods, such as are used in Computer Go. I am also impressed by some of the new applications, such as on autonomous helicopter flight and other robotics work.