Concealed Contributors to Result Quality — The Search Process of Ant Colony System

Stochastic solvers are researched primarily with the goal of providing ‘black box’ optimisation approaches for situations where the optimisation problem is too complex to model and therefore impossible to solve using a deterministic approach. Sometimes, however, problems or their instances have characteristics which interact with the solver in undocumented and unpredictable ways. This paper reviews some pertinent examples in the literature and provides an experiment which demonstrates that ant colony optimisation has arcane mechanisms which are partly responsible for results which are currently attributed to the pheromone-based learning.