Why adding more constraints makes a problem easier for hill-climbing algorithms: Analyzing landscapes of CSPs
It is well known that constraint satisfaction problems (CSPs) in the phase transition region are most difficult for complete search algorithms. On the other hand, for incomplete hill-climbing algorithms, problems in the phase transition region are more difficult than problems beyond the phase transition region, i.e., more constrained problems. This result seems somewhat unnatural since these more constrained problems have fewer solutions than the phase transition problems.
In this paper, we clarify the cause of this paradoxical phenomenon by exhaustively analyzing the state-space landscape of CSPs, which is formed by the evaluation values of states. The analyses show that by adding more constraints, while the number of solutions decreases, the number of local-minima also decreases, thus the number of states that are reachable to solutions increases. Furthermore, the analyses clarify that the decrease in local-minima is caused by a set of interconnected local-minima (basin) being divided into smaller regions by adding more constraints.
Unable to display preview. Download preview PDF.
- 1.Cheeseman, P., Kanefsky, B., and Taylor, W.: Where the really hard problems are, Proceedings of the Twelfth International Joint Conference on Artificial Intelligence (1991) 331–337Google Scholar
- 2.Clark, D. A., Frank, J., Gent, I. P., MacIntyre, E., Tomv, N., and Walsh, T.: Local Search and the Number of Solutions, Proceedings of the Second International Conference on Principles and Practice of Constraint Programming (Lecture Notes in Computer Science 1118), Springer-Verlag (1996) 119–133Google Scholar
- 3.Gent, I. P. and Walsh, T.: Towards an Understanding of Hill-climbing Procedures for SAT, Proceedings of the Eleventh National Conference on Artificial Intelligence (1993) 28–33Google Scholar
- 4.Hertz, A., Jaumard, B., and de Aragao, M. P.: Local optima topology for the k-coloring problem, Discrete Applied Mathematics, Vol. 49, (1994) 257–280Google Scholar
- 5.Hogg, T., Huberman, B. A., and Williams, C. P.: Phase transitions and the search problem, Artificial Intelligence, Vol. 81, No. 1–2, (1996) 1–16Google Scholar
- 6.Ishida, T.: Real-Time Bidirectional Search: Coordinated Problem Solving in Uncertain Situations, IEEE Transactions on Pattern Analysis and Machine Intelligence, Vol. 18, No. 6, (1996) 617–628Google Scholar
- 7.Mitchell, D., Selman, B., and Levesque, H.: Hard and Easy Distributions of SAT Problem, Proceedings of the Tenth National Conference on Artificial Intelligence (1992) 459–465Google Scholar
- 8.Morris, P.: The Breakout Method for Escaping From Local Minima, Proceedings of the Eleventh National Conference on Artificial Intelligence (1993) 40–45Google Scholar
- 9.Selman, B. and Kautz, H.: Domain-Independent Extensions to GSAT: Solving Large Structured Satisfiability Problems, Proceedings of the Thirteenth International Joint Conference on Artificial Intelligence (1993) 290–295Google Scholar
- 10.Selman, B., Levesque, H., and Mitchell, D.: A New Method for Solving Hard Satisfiability Problems, Proceedings of the Tenth National Conference on Artificial Intelligence (1992) 440–446Google Scholar