Promoting Search Diversity in Ant Colony Optimization with Stubborn Ants

In ant colony optimization (ACO) methods, including Ant System and MAX-M IN Ant System, each ant stochastically generates its candidate solution, in a given iteration, based on the same pheromone T and heuristic η information as every other ant. Stubborn ants is an ACO variation in which if an ant generates a particular candidate solution in a given iteration, then the components of that solution will have a higher probability of being selected in the candidate solution generated by that ant in the next iteration. In previous work, we evaluated this variation with the M M AS Ant System model and the Traveling Salesman Problem (TSP), and found that it can both improve solution quality and reduce execution-time. In this paper, we evaluate stubborn ants with Ranked Ant System, and find that performance also improves in terms of solution quality and execution time.