20113 Improvement of Local Search Performance by Direct Deformation of Landscape
Manabu Hasegawa, Yosuke Takahashi · The Proceedings of Conference of Kanto Branch · 2005
On the basis of the previous findings on the intensification characteristic in optimization by the Metropolis algorithm, the improvement of local search performance using direct deformation of landscape of cost function is studied by numerical experiment on a random instance of graph partitioning problem. As seen in the previous case of Euclidean traveling salesman problem, the optimization performance is maximized at the intermediate temperature which depends on the employed number of search steps. The maximum optimization performance seems to be improved by deformation whereby the search is guided to better basin through the pass over the lower hill of the original landscape.