Hyper-heuristics generate heuristics for problem classes

Meta-heuristics sample a search space, with quality dictated by an objective function. For any pair of metaheuristics, there is a pair of objective functions with identical performance. A similar statement can be made about problem classes (probability distributions over problem instances). The intuition behind this result is that a meta-heuristic can be viewed as a conditional probability over the search space and therefore this result can be considered as a conservation law. The contribution is an implication that meta-heuristics should be designed for a problem class. A natural solution to the ``metaheuristic design problem’ ’ is to employ generative hyper-heuristics to yield heuristics tailored to the problem class. Full Length Abstract. Meta-heuristics operate by sampling a search space of candidate solutions, with quality dictated by an objective function. By permutin