Solving partial constraint satisfaction problems using local search and abstraction

Abstract

Partial constraint satisfaction problems (PCSPs) were proposed by Freuder and Wallace to address some of the representational difficulties with traditional constraint satisfaction techniques. However, the reasoning method of their proposal was limited to traditional backtracking based algorithms. In this paper, we extend the PCSP model by associating it with a local search algorithm, which has found great successes in solving many large scale problems in the past. Furthermore, we extend the combined model to incorporate abstract problem solving, and show that the extended model has not only the advantages of both PCSP and local search, but also a number of new features useful for scheduling applications. We demonstrate the feasibility of our approach by an application to a university course scheduling domain.