Effect of DisCSP variable-ordering heuristics in scale-free networks
PRIMA'10 Proceedings of the 13th international conference on Principles and Practice of Multi-Agent Systems
Effect of DisCSP variable-ordering heuristics in scale-free networks
Multiagent and Grid Systems - Principles and Practice of Multi-Agent Systems
| Hi-index | 0.00 |
Many complex real-world systems can be modeled using a graphicalstructure such as a constraint network. If the properties of sucha structure can be exploited, many challenging computationaltasks can have good typical-case runtimes even if theyare theoretically intractable in general. In this paper we show that many real-world constraint networks induce binary networks that share a common underlying structural characterisation; namely, that their degree distributionsexhibit preferential attachment. We report on a novel constraint network generator for randomconstraint networks that have a scale-free macrostructure. This scale-free generator is based on the well known Barabasi-Albert preferential attachment model. Using this model we demonstrate that real-world constraint networksexhibit degree distributions that are more like those found inscale-free graphs. We also show that the effect of standard degree-based search heuristics on real-world problems exhibiting power-law degree distributions is greater than problems with a uniform random structure. We also show that the backdoor sizes for preferentially attached constraint networks are smaller than those of uniform randomproblems. This paper provides a novel basis for studying realistic constraintmodels.