« Learning to solve complex problems | Main | French teams getting together »

Comments

Bianca Zadrozny

The control problem is exactly the kind of problem that reinforcement learning addresses. In my thesis (available at http://www.cse.ucsd.edu/~zadrozny/ ), I also address this problem in a setup different from traditional reinforcement learning. I assume that I have a batch of examples of the kind (s,a,l) where l is the loss of executing action a in state s (in your example s=S, a=G and l=T). In reinforcement learning instead of the batch of examples we assume we can interact with the environment to collect data.

The comments to this entry are closed.