Graph coloring by multiagent fusion search

A multiagent fusion search is presented for the graph coloring problem. In this method, each of agents performs the fusion search, involving a local search working in a primary exploitation role and a recombination search in a navigation role, with extremely limited memory and interacts with others through a decentralized protocol, thus agents are able to explore in parallel as well as to achieve a collective performance. As the knowledge components implemented with available structural information and in formalized forms, the Quasi-Tabu local search and grouping-based recombination rules are especially useful in addressing neutrality and ruggedness of the problem landscape. The new method has been tested on some hard benchmark graphs, and has been shown competitive in comparison with several existing algorithms. In addition, the method provides new lower bound solutions when applied to two large graphs. Some search characteristics of the proposed method is also discussed.