Towards ‘Metaheuristics in the Large’

Abstract

There is a pressing need for a higher-level architectural per- spective in metaheuristics research. This article proposes a purely functional collection of component signatures as a basis for the scalable and automatic construction of meta- heuristics. We claim that this is an important step for sci- entific progress because: i). It is increasingly accepted that newly-proposed meta- heuristics should be grounded in terms of well-defined frameworks and components. Standardized descrip- tions help to distinguish novelty from minor variation. ii). Greater reproducibility is needed, particularly to facil- itate comparison with the state-of-the-art. iii). Interoperable descriptions are a pre-requisite for a data model supporting large-scale knowledge discovery across frameworks and problems. A key obstacle is that metaheuristic components suffer from an intrinsic lack of modularity, so we present some design op- tions for dealing with this and use this to provide a roadmap for addressing the above issues.