Bounded-Resource Reasoning as (Strong or Classical) Planning

摘要

To appropriately configure agents so as to avoid resource exhaustion, it is necessary to determine the minimum resource (time & memory) requirements necessary to solve reasoning problems. In this paper we show how the problem of reasoning under bounded resources can be recast as a planning problem. Focusing on propositional reasoning, we propose different recasting styles, which are equally interesting, since they require solving different classes of planning problems, and allow representing different reasoner architectures. We implement our approach by automatically encoding problems for the MBP planner. Our experimental results demonstrate that even simple problems can give rise to non-trivial (and often counter intuitive) time and memory saving strategies.