Tabu Searchのグラフ分割問題への適用と実験的解析

In this paper, we report on an application of tabu search to the graph partitioning problem which has applications on circuit board wiring and program segmentation. We discuss how to adapt tabu search to the graph partitioning problem and compare the performance with simulated annealing, another variant of local search incorporating randomized technique. Numerical experiments show that our algorithm dominates the simulated annealing algorithm in accuracy of solutions and speed on both uniform and geometric instances. In particular, our tabu search implementation works much better than the simulated annealing algorithm on structured (geometric) instances. We also investigate how to tune up our implementation and to optimize the various parameters via extensive numerical experiments.