Efficiently identifying pareto solutions when objective values change
Fieldsend, Jonathan E.
Everson, Richard M.
In many multi-objective problems the objective values assigned to a particular design can change during the course of an optimisation. This may be due to dynamic changes in the problem itself, or updates to estimated objectives in noisy problems. In these situations, designs which are non-dominated at one time step may become dominated later not just because a new and better solution has been found, but because the existing solution's performance has degraded. Likewise, a dominated solution may later be identified as non-dominated because its objectives have comparatively improved. We propose management algorithms based on recording single “guardian dominators" for each solution which allow rapid discovery and updating of the non-dominated subset of solutions evaluated by an optimiser. We examine the computational complexity of our proposed approach, and compare the performance of different ways of selecting the guardian dominators.
Copyright © 2014 ACM
2014 Conference on Genetic and Evolutionary Computation (GECCO ’14), Vancouver, BC, Canada, 12-16 July 2014
Proceedings of the 2014 Conference on Genetic and Evolutionary Computation, pp. 605 - 612