Towards the development of self-ant systems
Jorge Tavares, Francisco Baptista Pereira · 2011
We propose a computational framework for the self-generation of components used by an Ant Colony Optimization algorithm. The approach relies on Strongly Typed Genetic Programming to automatically seek for effective update pheromone strategies. Best evolved strategies are then inserted in an Ant Colony Algorithm used to find good quality solutions for the Quadratic Assignment Problem. Results reveal that evolved update rules are competitive with human designed variants and can be effectively reused on different instances of the same problem. Moreover, we investigate the possibility of evolving general strategies that can be used across different optimization problems.