Memetic algorithm with strategic controller for the maximum clique problem

  • Authors:

  • Kengo Katayama;Akinori Kohmura;Keiko Kohmoto;Hideo Minamihara

  • Affiliations:

  • Okayama University of Science, Kita-ku, Okayama, Japan;Okayama University of Science, Kita-ku, Okayama, Japan;Kinki University, Kinokawa-city, Wakayama, Japan;Okayama University of Science, Kita-ku, Okayama, Japan

  • Venue:
  • Proceedings of the 2011 ACM Symposium on Applied Computing
  • Year:
  • 2011

Quantified Score

Hi-index 0.00

Visualization

Abstract

Most of standard evolutionary algorithms consist of a mutation, a crossover, a selection and often a local search. Each of these operators is specifically designed for a combinatorial optimization problem. These can be considered as tools for the optimization searches, and the interplay between them in the searches is not apparently controlled in many cases. In this paper, we present a flexible control method, called Strategic Controller (SC), for multiple mutation methods equipped in a memetic algorithm (MA) for the maximum clique problem (MCP). The SC is used to choose a suitable method from the candidate mutations. To perform an adaptive search, the SC evaluates each mutation method based on the contribution information which is served as novel "memes" for the mutations in the MA. To achieve the SC, we apply the idea of analytic hierarchy process. Although standard MAs have a population of multiple solutions as memes usually, a single solution is used in our MA. We evaluated the performance of MA with SC (MA-SC) on DIMACS benchmark graphs of the MCP. The results showed that MA-SC is capable of finding comprehensive solutions through comparisons with MAs in which each mutation is used. Moreover, we observed that it is highly effective particularly for hardest graphs in the benchmark set in comparisons with recent metaheuristics to the MCP.