The Same-Decision Probability: A New Tool for Decision Making

When using graphical models for decision making, a fundamental question is whether one is ready to make a decision (stopping criteria), and if not, what observations should be made to better prepare for a decision (selection criteria). In this paper, we review the notions of entropy and expected utility, which are commonly used for this purpose, and contrast them with a newly introduced notion, called the same-decision probability, which can be used both as a stopping criteria for making decisions and as a selection criteria for choosing additional observations. Furthermore, we show that computing the samedecision probability lies in the same complexity class as a general expectation computation problem that is applicable to a wide variety of queries in graphical models, including the computation of non-myopic value of information.