An analysis of a post-selection mechanism for handling stochasticity in tuning numerical parameters of MAX–MIN Ant System

Many nature-inspired optimization techniques have a number of parameters, and automated algorithm tuning has shown to be instrumental for deriving high-performing algorithms. In previous work, we have studied state-of-the-art derivative-free continuous optimization algorithms together with techniques for handling the stochasticity arising in the tuning problems to fine-tune numerical parameters. In this work, we extend this earlier work and provide an in-depth investigation on post-selection, a simple mechanism for stochasticity handling. In addition, we study the behavior of the tuning algorithms with respect to different amount of tuning budget. Empirical results confirm the effectiveness and robustness of the post-selection mechanism across different levels of tuning budget.