Clustering of Local Optima in Combinatorial Fitness Landscapes
This work provides insights into problem space structure for combinatorial optimization, which could inform heuristic search strategies, but it is incremental as it builds on existing network models.
The study analyzed the distribution of local optima in quadratic assignment problem instances using local optima networks, finding that real-like instances have a clear modular structure while random uniform instances are less clusterable.
Using the recently proposed model of combinatorial landscapes: local optima networks, we study the distribution of local optima in two classes of instances of the quadratic assignment problem. Our results indicate that the two problem instance classes give rise to very different configuration spaces. For the so-called real-like class, the optima networks possess a clear modular structure, while the networks belonging to the class of random uniform instances are less well partitionable into clusters. We briefly discuss the consequences of the findings for heuristically searching the corresponding problem spaces.