Q-learning on Games of Imperfect Information

We applied Q-learning, the process of learning the value of state-action pairs, to games where the full state is not surfaced to the decision maker. Our full results are described here, but, in summary, we determined that a Q-learning agent could eventually understand the likelihood of various world states given the surfaced state and act accordingly.