Compiling constraint satisfaction problems

Many tasks requiring intelligence, in particular scheduling and planning, must be solved under time constraints. This is diicult to achieve because of the combinato-rial nature of such tasks. While search heuristics can give good average performance, they cannot give any peformance guarantees for a particular instance. Fortunately, the tasks are often very similar. Therefore, compiling partial solutions is one way in which better performance guarantees for on-line problem solving could be achieved. We consider constraint satisfaction as a general paradigm and describe compilation techniques. General tasks are deened by incomplete CSPs from which instances are generated by adding more constraints. For any such general task, compilation builds a structure which represents all its solutions. In order to represent the space in a compact form, it exploits clustering and interchangeability techniques. Search for solutions can then be limited to a usually much smaller, precomputed space. When search criteria involve only single variables, solutions can be guaranteed to be found in linear time in the size of the compiled structure.