Computationally-Optimal Real-Resource Strategies

This paper focuses on managing the cost of de­ liberation before action. In many problems, the cost and the resource consumption of the delib­ eration phase cannot be ignored, and the overall quality of the solution reflects the costs incurred and the resources consumed in deliberation as well as the cost and benefit of execution. A feasible strategy that minimizes the total cost is termed computationally-optimal. For a situation where a number of independent, uninterruptible meth­ ods are available to solve the problem, we develop a pseudopolynomial-time algorithm to construct generate-and-test computationally-optimal strate­ gies. Stochastic Dynamic Programming is used to solve the problem that is shown to be NP-complete and the results address problems occurring in auto­ matic emergency-response systems, design automa­ tion, query optimization, destructive testing, and other areas characterized by significant computa­ tional costs or limited deliberation resources.