Estimating the quality of initial populations in multi-objective evolutionary algorithms
Tobias Benecke, Sanaz Mostaghim · Proceedings of the Genetic and Evolutionary Computation Conference Companion · 2022
The initial population is the starting point of each evolutionary algorithm (EA) and is widely understood to be an important factor when it comes to the overall performance of the algorithm. In this paper, we explore the possibility of estimating the quality of the initial population in multi-objective evolutionary algorithms (MOEAs) without using expensive benchmarking. While there are already well established metrics for measuring the performance of a MOEA, they are often calculated only on the non-dominated solutions and show drawbacks when applied as a measure for the first generation. Here, we discuss the challenges for estimating the quality of the initial population and propose a new metric dedicated for this purpose. Finally, we evaluate the accuracy of our metric against a set of benchmarking problems.