Interactive tool for visualizing the comprehensive performance of evolutionary multi-objective algorithms applied to problems with two or three objectives
Michał K. Tomczyk, Miłosz Kadziński · Proceedings of the Genetic and Evolutionary Computation Conference Companion · 2024
The performance of evolutionary algorithms for multi-objective optimization is typically assessed by considering only the final populations. This is troublesome for the following reasons. First, it ignores most solutions constructed throughout the evolution, thus not portraying overall progression. Second, evolutionary methods are susceptible to randomness. Therefore, the results obtained from a single test run are unreliable. Third, these methods are complex algorithms that can be evaluated from many perspectives, not just by assessing the qualities of the final solutions they present. Overcoming these issues motivated the development of our novel visualization tool that accounts for robustness in assessment. It supports both 2D and 3D visualization, and, in the case of the latter, it introduces interactivity, allowing the inspection of presented results in a manner most suited to the user. The developed software is an inherent part of this paper and can be downloaded and re-used by researchers to foster their research.