Search plays a central role in solving a wide range of problems in many fields of computer science and, in particular, artificial intelligence (AI). For example, recent successes in AI disciplines like planning and scheduling, game playing, and constraint programming are in big part due to the development of effective search techniques for exploring huge search spaces. As a way of tackling the inherited combinatorial complexity of such problems, these methods use intelligent heuristics and various other enh ancements to help them focus the exploration effort and prune the search space. Developing such intelligent search-control heuristics often requires much effort and human-intervention for providing expert-level domain knowledge. There is consequently incre ased interest in search techniques that gradually learn to improve their efficiency automatically by dynamically adapting their exploration strategy based on experience gained from current or previous problem solving episodes. Unfortunately, existing rule-based search-control learning mechanisms do not work well for exploring huge search spaces. The main object of this proposal is to develop new adaptive search control techniques especially designed for real-time heuristic search in huge search spaces, with application in domains like game-playing, problem-solving, automated planning, and network routing. This objective will be achieved by building up on recent work on search-control in adversary search domain; ideas from there will be generalized to be app licable in non-adversary domains.Automatic learning of search control in heuristic search has the potential of improving both the efficiency and the decision quality of heuristic search solvers, allowing them to solve larger problems than they are capable of today. This is important in the technical information society we live in today, where software applications need to handle ever increasing amount of data.
Fields of science
Call for proposal
See other projects for this call