Empirical linkage learning

Multi-Objective parameter-less population pyramid for solving industrial process planning problems

“…The methods that employ linkage learning are particularly effective in the optimisation of problems characterised by such solution spaces. This observation applies to both: theoretical [20,21,19,22] and practical problems [1,23,24]. It is also shown that methods employing linkage learning may significantly outperform the other that do not use such techniques [20,25,19,24].…”

“…MO-GOMEA is based on the concept of the Linkage Tree Genetic Algorithm (LTGA) [22,26,1]. LTGA is a Genetic Algorithm (GA) that employs linkage learning techniques [21,27,22,28] to improve its effectiveness. Similarly to LTGA in the single-objective domains, MO-GOMEA has significantly outperformed competing methods (including NSGA-II and MOEA/D) in multi-objective optimisation [10,19].…”

“…They show that if the problem structure is complex (there are many gene dependencies) [31,32], it is favourable to learn linkage during the method execution rather than obtaining the linkage in the pre-optimisation step. Finally, in [21], the authors show that to assure the method's effectiveness, the linkage should be of high quality, but it also should be diverse. To obtain this, they propose a method that utilises a multipopulation approach.…”

“…To obtain this, they propose a method that utilises a multipopulation approach. Note that the multi-population approaches are usually employed to increase population diversity [23,20,24], but they may also be useful in obtaining a diverse linkage [21]. Note that the lack of linkage learning diversity may be a likely reason for a poor performance of LTGA shown in [33].…”

“…The research considering the influence of linkage quality on the methods' performance is in its early stage. For instance, it requires further investigation of the reasons why linkage diversity is important to effectively solve problems with complex structure (including so-called overlapping building blocks, which is a typical feature of practical problems) [21]. Nevertheless, the objective of this paper is to use the conclusions of the research that has been already made in this area, and to apply these conclusions as intuitions that shall guide us to proposing an effective method for solving the MOBCPP problem.…”