Metaheuristics provide the means to approximately solve complex optimisation problems when exact optimisers cannot be utilised. This led to an explosion in the number of novel metaheuristics, most of them metaphor-based, using nature as a source of inspiration. Thus, keeping track of their capabilities and innovative components is an increasingly difficult task. This can be resolved by an exhaustive classification system. Trying to classify metaheuristics is common in research, but no consensus on a classification system and the necessary criteria has been established so far. Furthermore, a proposed classification system can not be deemed complete if inherently different metaheuristics are assigned to the same class by the system. In this paper we provide the basis for a new comprehensive classification system for metaheuristics. We first summarise and discuss previous classification attempts and the utilised criteria. Then we present a multi-level architecture and suitable criteria for the task of classifying metaheuristics. A classification system of this kind can solve three main problems when applied to metaheuristics: organise the huge set of existing metaheuristics, clarify the innovation in novel metaheuristics and identify metaheuristics suitable to solve specific optimisation tasks.
Automated prediction systems based on machine learning (ML) are employed in practical applications with increasing frequency and stakeholders demand explanations of their decisions. ML algorithms that learn accurate sets of rules, such as learning classifier systems (LCSs), produce transparent and human-readable models by design. However, whether such models can be effectively used, both for predictions and analyses, strongly relies on the optimal placement and selection of rules (in ML this task is known as model selection). In this article, we broaden a previous analysis on a variety of techniques to efficiently place good rules within the search space based on their local prediction errors as well as their generality. This investigation is done within a specific pre-existing LCS, named SupRB, where the placement of rules and the selection of good subsets of rules are strictly separated—in contrast to other LCSs where these tasks sometimes blend. We compare two baselines, random search and $$(1, \lambda )$$ -evolution strategy (ES), with six novelty search variants: three novelty-/fitness weighing variants and for each of those two differing approaches on the usage of the archiving mechanism. We find that random search is not sufficient and sensible criteria, i.e., error and generality, are indeed needed. However, we cannot confirm that the more complicated-to-explain novelty search variants would provide better results than $$(1, \lambda )$$ -ES which allows a good balance between low error and low complexity in the resulting models.
We present MAHF, a Rust framework for the modular construction and subsequent evaluation of evolutionary algorithms, but also any other metaheuristic framework, including non-population-based and constructive approaches. We achieve high modularity and flexibility by splitting algorithms into components with a uniform interface and allowing communication through a shared blackboard. Nevertheless, MAHF is aimed at being easy to use and adapt to the specific purposes of different practitioners. To this end, this paper focuses on providing a general description of the design of MAHF before illustrating its application with a variety of different use cases, ranging from simple extension of the set of implemented components and the subsequent construction of algorithms not present within the framework to hybridization approaches, which are often difficult to realize in specialized software frameworks. By providing these comprehensive examples, we aim to encourage others to utilize MAHF for their needs, evaluate its effectiveness, and improve upon its application.
Practitioners frequently encounter the challenge of selecting the best optimization algorithm from a pool of options. However, why not, rather than selecting a single algorithm, let evolution determine the optimal combination of all algorithms? In this paper, we present an approach to algorithm design inspired by a well-known traditional method for coarse-grained hybridization: the heterogeneous island model. Our hyper-heuristic framework represents island models as graphs and identifies optimal island topologies and parameters for specific sets of problem instances. Since the framework operates at the level of metaheuristic algorithms rather than components and incorporates a configuration mechanism directly into the search, it combines concepts from algorithm design, selection, and configuration. The proposed framework is investigated on 24 training sets of varying difficulty and demonstrates its ability to discover complex hybrids. A post-evaluation on real-world constrained optimization problems shows a significant improvement over the algorithms on their own. These results suggest that it is a promising way to design hybrid metaheuristics with minimal manual intervention, given representative training instances, a set of optimization algorithms, and sufficient computational resources.