Fitness-Distance Correlation and Solution-Guided Multi-point Constructive Search for CSPs

Solution-Guided Multi-Point Constructive Search (SGMPCS) is a complete, constructive search technique that has been shown to out-perform standard constructive search techniques on a number of constraint optimization and constraint satisfaction problems. In this paper, we perform a case study of the application of SGMPCS to a constraint satisfaction model of the multi-dimensional knapsack problem. We show that SGMPCS performs poorly. We then develop a descriptive model of its performance using fitness-distance analysis. It is demonstrated that SGMPCS search performance is partially dependent upon the correlation between the heuristic evaluation of the guiding solutions and their distance to the nearest satisfying solution. This is the first work to develop a descriptive model of SGMPCS search behavior. The descriptive model points to a clear direction in improving the performance of constructive search for constraint satisfaction problems: the development of heuristic evaluations for pa...

• Type: Conference
• Year: 2008
• Where: CPAIOR
• Authors: Ivan Heckman, J. Christopher Beck
• Keywords: Constraint Satisfaction, Constructive Search, Constructive Search Techniques, CPAIOR 2008, Operations Research