Tuning MAX–MIN Ant System with off-line and on-line methods

In this paper, we study two approaches for tuning parameters of optimization algorithms: off-line and on-line tuning. The goal of tuning is to find an appropriate configuration of an algorithm, where a configuration is a specific setting of all parameters. In off-line tuning, the appropriate configuration is chosen after a preliminary phase. In on-line tuning, the configuration might be changed during the solution process to adapt to the instance being solved and possibly to the state of the search. Parameter tuning is very important in swarm intelligence: the performance of many techniques is strongly influenced by the configuration chosen. In this paper, we study the performance of MAX–MIN Ant System when its parameters are tuned off-line and when they are tuned on-line. We consider one off-line and five on-line methods, and we tackle two combinatorial optimization problems. The results show that off-line tuning generally outperforms on-line tuning. In some cases, on-line tuning achieves better results than off-line tuning, but the best on-line method is not always the same.