Over the last decade, AI researchers have investigated flexible inferential procedures and representations that allow reasoning systems to gracefully trade off one or more dimensions of the quality of inferred results for the quantity of time or memory required to generate the results. Methods for monitoring and controlling flexible procedures hold opportunity for endowing intelligent systems with the ability to tailor inference dynamically to specific limitations or variations in available computational resources, depending on the situation or environment.
Research on flexible procedures and the control of computational tradeoffs under bounded resources has been referred to in a variety of ways depending on specific details of the procedures and the application area, including flexible computation, anytime algorithms, imprecise computation, designto-time scheduling, memory-bounded search, and resource-bounded reasoning. This special issue of the journal Arti'cial Intelligence will bring together articles describing effective solutions to challenges with the development and control of flexible procedures, including problems with the composition, monitoring, and guidance of inference under limited resources. In addition to papers on principles for characterizing and handling computational tradeoffs under bounded resources, we invite studies in application areas such as heuristic search, constraint satisfaction, probabilistic inference, planning and scheduling, signal interpretation, medical diagnosis and treatment, and intelligent information retrieval.
Topics of interest include: l computational tradeoffs in inference, planning, and search, l flexible representations of knowledge, problem instances, l representation and measurement of computational tradeoffs, l methods for learning about performance and tradeoffs, l dependency of performance on details of problem instances, l flexibility in problem solving procedures, 0 real-time monitoring of solution quality, l strategies for allocating resources among reasoning subproblems, l partitioning resources between metalevel and object-level reasoning,