Preference-Driven Co-evolutionary Algorithms Show Promise for Many-Objective Optimisation
- Cite this paper as:
- Purshouse R.C., Jalbă C., Fleming P.J. (2011) Preference-Driven Co-evolutionary Algorithms Show Promise for Many-Objective Optimisation. In: Takahashi R.H.C., Deb K., Wanner E.F., Greco S. (eds) Evolutionary Multi-Criterion Optimization. EMO 2011. Lecture Notes in Computer Science, vol 6576. Springer, Berlin, Heidelberg
The simultaneous optimisation of four or more conflicting objectives is now recognised as a challenge for evolutionary algorithms seeking to obtain full representations of trade-off surfaces for the purposes of a posteriori decision-making. Whilst there is evidence that some approaches can outperform both random search and standard Pareto-based methods, best-in-class algorithms have yet to be identified. We consider the concept of co-evolving a population of decision-maker preferences as a basis for determining the fitness of competing candidate solutions. The concept is realised using an existing co-evolutionary approach based on goal vectors. We compare this approach and a variant to three realistic alternatives, within a common optimiser framework. The empirical analysis follows current best practice in the field. As the number of objectives is increased, the preference-driven co-evolutionary approaches tend to outperform the alternatives, according to the hypervolume indicator, and so make a strong claim for further attention in many-objective studies.
Keywordsmany-objective optimisation co-evolution comparative study
Unable to display preview. Download preview PDF.