When Model Bias Is Stronger than Selection Pressure

We investigate the influence of model bias in model-based search. As an example we choose Ant Colony Optimization as a well-known model-based search algorithm. We present the effect of two different pheromone models for an Ant Colony Optimization algorithm to tackle a general scheduling problem. The results show that a pheromone model can introduce a strong bias toward certain regions of the search space, stronger than the selection pressure introduced by the updating rule for the model. This potentially leads to an algorithm where over time the probability to produce good quality solutions decreases.