Clustering for Disconnected Solution Sets of Numerical CSPs

This paper considers the issue of preprocessing the output of interval-based solvers for further exploitations when solving numerical CSPs with continuum of solutions. Most interval-based solvers cover the solution sets of such problems with a large collection of boxes. This makes it difficult to exploit their results for other purposes than simple querying. For many practical problems, it is highly desirable to run more complex queries on the representations of the solution set. We propose to use clustering techniques to reduce the number of boxes produced by interval-based solvers, while providing some main characteristics of the solution set. Four new algorithms based on clustering are proposed.