An adaptive divide-and-conquer methodology for evolutionary multi-criterion optimisation

Abstract. Improved sample-based trade-off surface representations for large numbers of performance criteria can be achieved by dividing the global problem into groups of independent, parallel sub-problems, where possible. This paper describes a progressive criterion-space decomposition methodology for evolutionary optimisers, which uses concepts from parallel evolutionary algorithms and nonparametric statistics. The method is evaluated both quantitatively and qualitatively using a rigorous experimental framework. Proof-of-principle results confirm the potential of the adaptive divide-and-conquer strategy.