Hyper-Heuristics
Also known as: Heuristic of Heuristics, Algorithm Selection Hyper-Heuristic, Selection Hyper-Heuristic, Hiyer-Sezgisel
Hyper-heuristics are high-level methodologies that search over a space of heuristics rather than directly over the space of solutions. Introduced systematically by Burke et al. (2013) in their landmark survey, hyper-heuristics operate by selecting or generating low-level heuristics to solve hard combinatorial optimisation and search problems, aiming to automate the design of optimisation algorithms across diverse problem domains without requiring deep problem-specific knowledge.
Key highlights
- Domain-generality: the high-level search component is reusable across different problem classes without modification.
- Automation of algorithm design: reduces reliance on expert knowledge to construct problem-specific metaheuristics.
- Adaptability: learning-based selection mechanisms allow the framework to improve performance during a single run.
- Robustness: maintains competitive solution quality across diverse problem instances without extensive parameter tuning.
Intuition
This section is available to Pro members. Upgrade to Pro
How it works
This section is available to Pro members. Upgrade to Pro
When to use it
Hyper-heuristics are appropriate when you face a combinatorial optimisation problem but lack the expertise or time to design a tailored metaheuristic, or when you need a single framework reusable across multiple related problem instances. They are particularly suited to timetabling, scheduling, bin-packing, and routing. Key assumptions include the availability of a set of low-level heuristics and a solution evaluation function. Limitations include potentially lower peak performance than highly tuned problem-specific algorithms and sensitivity to the quality of the provided heuristic pool.
Strengths & limitations
- Domain-generality: the high-level search component is reusable across different problem classes without modification.
- Automation of algorithm design: reduces reliance on expert knowledge to construct problem-specific metaheuristics.
- Adaptability: learning-based selection mechanisms allow the framework to improve performance during a single run.
- Robustness: maintains competitive solution quality across diverse problem instances without extensive parameter tuning.
- Peak performance is typically below that of a highly customised, problem-specific metaheuristic on any single problem.
- Quality of results depends heavily on the richness and diversity of the supplied low-level heuristic pool.
- Heuristic generation hyper-heuristics (e.g., grammar-based) introduce high computational overhead.
- Theoretical convergence guarantees are largely absent, making formal performance analysis difficult.
Common pitfalls
This section is available to Pro members. Upgrade to Pro
Applications
This section is available to Pro members. Upgrade to Pro
Frequently asked
What distinguishes a hyper-heuristic from a metaheuristic?
A metaheuristic directly searches the problem solution space using problem-specific operators. A hyper-heuristic operates one level above: it searches the space of heuristics, selecting or generating which operator to apply next. This separation provides domain-independence — the same hyper-heuristic framework can be reused across entirely different problem types by simply swapping the low-level heuristic pool.
Can hyper-heuristics generate new heuristics, or only select existing ones?
Both are possible. Selection hyper-heuristics choose from a fixed pool of predefined low-level heuristics at each decision point. Generation hyper-heuristics (e.g., using genetic programming or grammar-based evolution) automatically construct new heuristics during the search. Generation approaches offer higher flexibility but substantially greater computational cost and implementation complexity.
How should I construct the low-level heuristic pool?
The pool should contain diverse, complementary operators that collectively cover the main structural features of the problem. Typical choices include local search moves (e.g., swap, insert, invert), construction heuristics, and perturbation operators. Redundant or nearly identical heuristics waste selection capacity. A domain expert should supply and validate the pool, even though the hyper-heuristic layer itself remains domain-independent.
Sources
- 1.Burke, E. K., et al. (2013). Hyper-heuristics: A survey of the state of the art. Journal of the Operational Research Society, 64(12), 1695–1724.
You have read it. What now?
Cite this page
ScholarGate. (2026, June 2). Hyper-Heuristics. ScholarGate. https://scholargate.app/optimization/hyper-heuristics