An Expedition to Multimodal Multi-Objective Optimization Landscapes
Kerschke Pascal, Grimme Christian
The research in evolutionary multi-objective optimization islargely missing a notion of functional landscapes, which could enablea visual understanding of multimodal multi-objective landscapes andtheir characteristics by connecting decision and objective space. Thisconsequently leads to the negligence of decision space in most algorithmicapproaches and an almost complete lack of Exploratory LandscapeAnalysis (ELA) tools. This paper dares a rst step into this unexplored eld based on gradient properties of the multi-objective landscape. Fora rst time, basins of attraction and superpositions of local optima arevisualized and thereby made intuitively accessible. With this work, wehope to highlight the importance of detailed decision space analysis inmulti-objective optimization and to stimulate further research in thatdirection.
Multi-objective evolutionary optimization; multimodality; fitness landscapes; basins of attraction; visualization; qualitative analysis