Exploitation and Exploration vs Difficulty
Maurice Clerc · 2019
Essentially, the sampling strategies of an optimizer are responsible for the latter assuming whether a problem is easy or difficult. For example, for a purely greedy algorithm, a unimodal problem is very easy and a multimodal problem is very difficult as soon as the basin of attraction of the global minimum is significantly smaller than that of a local minimum. The classic mantra that can be found in many articles and books is that there must be a “balance” between exploitation and exploration (sometimes referred to as intensification and diversification). As such, definitions sufficiently accurate to calculate these two quantities and observe the evolution of their ratio during the iterations are rarely given. To clearly bring forward the influence of the exploitation/exploration ratio, this chapter explores rigorous definitions of these two concepts, making it possible to measure them.