Tracing Genome Influence in Multi-Objective Evolutionary Algorithms

Multi-objective evolutionary algorithms (MOEAs) are well established in their application and feature an active field of research. However, when it comes to developing, testing and comparing these algorithms, usually only their performance in the objective space is considered. This results in a lack of understanding when it comes to the dynamics of MOEAs in the search space. The focus of this work is to better understand the behavior of MOEAs by studying its population dynamics. We aim towards finding out which individuals from the initial population contribute in the search towards the Pareto-front. For this, we use the traceable evolutionary algorithm, which allows us to trace the genetic materials and their influence over the evolutionary process. Defining good genetic material for algorithms with more than one objective, however, proves difficult compared to the single-objective case. We evaluate the effects of different known optimal solutions both on the performance and differences in search behavior of MOEAs. Furthermore, we evaluate the population dynamics for different test problems, algorithms, and seed individuals on different locations on the Pareto-front.