Learning Multiple Search, Utility, And Goal Parameters For The Game RISK

In dynamic planning and execution problems the measure of utility is the value of the reward one seeks combined with the probability of achieving that reward. However, in a complex stochastic environment, there are a number of other concerns when calculating the true utility of planning ahead and achieving predicted results. Three additional factors that can be considered in measuring a broader, more versatile, utility metric are: (1) the expected value may produce more risk than desired; (2) the temporal cost of planning; and (3) a more comprehensive consideration of the probability of successful completion of a plan. The correct application of these parameters is not fixed and may depend on the application. In this paper, we present a framework for learning these parameters with the inclusion of reward and solution search parameters to formulate a truer measure of success. We also present a specific example of learning these parameters for the game RISK.