The effect of the number of ants parameter in the ACOR algorithm: A run-time profiling study

The number of ants is an important controlling parameter in the ACO _R algorithm. We present a profiling study in which we run ACO _R a large number of times on a suite of popular benchmark continuous-domain optimization problems, for the purpose of trying to understand how the algorithm's behavior changes with the number of ants. We consider the percentage of newly constructed solutions that find a place in the population and consider where in the (sorted) population those solutions are placed. We also consider the percentage of solutions that would secure a place in the population if a single ant was used, but potentially may not find a place in the population if multiple ants were used. We present an argument that the choice of the number of ants can be framed in terms of the exploitation-exploration dilemma, with a smaller number of ants favoring exploration.