The Danger of Metaphors for Metaheuristic Design
Stephen Y. Chen, Shehnaz Islam, Shanshan Lao · 2023
The design of metaheuristics for numerical and combinatorial optimization suffers from an explosion of techniques. Many of these techniques can be described by the same pseudo-code in which the method of generating new search solutions changes, but the overall technique follows a standard Evolutionary Computation or Swarm Intelligence framework. The benefit of new metaphors for the design of metaheuristics has been called into question by prominent members of the research community. However, the benefit of metaphors in their entirety remains largely unquestioned. A new experimental analysis of a foundational metaheuristic demonstrates behaviours that are distinctly different from its metaphor-based inspiration. It appears that detailed analysis has been superseded by the general assumption that an algorithm will behave in accordance to the metaphor it has been designed to mimic. The ability of a metaphor-based design to misguide the analysis and observation of its operation represents a key danger in the use of metaphors to design metaheuristics.