In this period I will research existing work on techniques for evaluation of problems involving large decision trees. In addition, I will study and become acquainted with the existing research on the application of decision-theoretic control techniques to large search and game-playing problems.
This work will involve talking to experts in application of AI to real-world decision-making problems, such as Steve Pauker at NEMC. In these discussions I hope to learn more about the nature of the large decision problems encountered, to learn about the problems with the use of existing techniques, and to determine to what degree my own approach would be appropriate for various kinds of real-world decision problems.
I also plan to look into the computer-game playing research relevant to one-player games with chance, to determine whether partial evaluation of decision trees would be a profitable approach in that domain as well, and if so, to study other techniques against which to make objective comparisons of performance. Also, there is the possibility that I may find some techniques in the solitaire-playing literature that could be applied to some real-world decision problems.
Also, I will study in detail the metareasoning work of Russell &Wefald, Horvitz, Hansson &Mayer, and others, to thoroughly learn how to correctly apply decision theory to the selection and control of computational actions when searching for approximate solutions to hard problems.