ParetoLens: A Visual Analytics Framework for Exploring Solution Sets of Multi-objective Evolutionary Algorithms

TL;DR

ParetoLens addresses the challenge of analyzing high-dimensional solution sets from multi-objective evolutionary algorithms by introducing a modular visual analytics framework. It combines two projection views, a parallel coordinates/PCP view, density-based reference maps, and cross-component coordination to support interactive exploration, quality assessment, and decision-space correlations. The approach is validated through three benchmark case studies, data-speed evaluations, and expert interviews, demonstrating improved pattern discovery and hypothesis testing in MOEAs. The work lays a foundation for scalable, domain-agnostic MOEA analysis with potential for extension to multi-set comparisons and domain-specific visualizations.

Abstract

In the domain of multi-objective optimization, evolutionary algorithms are distinguished by their capability to generate a diverse population of solutions that navigate the trade-offs inherent among competing objectives. This has catalyzed the ascension of evolutionary multi-objective optimization (EMO) as a prevalent approach. Despite the effectiveness of the EMO paradigm, the analysis of resultant solution sets presents considerable challenges. This is primarily attributed to the high-dimensional nature of the data and the constraints imposed by static visualization methods, which frequently culminate in visual clutter and impede interactive exploratory analysis. To address these challenges, this paper introduces ParetoLens, a visual analytics framework specifically tailored to enhance the inspection and exploration of solution sets derived from the multi-objective evolutionary algorithms. Utilizing a modularized, algorithm-agnostic design, ParetoLens enables a detailed inspection of solution distributions in both decision and objective spaces through a suite of interactive visual representations. This approach not only mitigates the issues associated with static visualizations but also supports a more nuanced and flexible analysis process. The usability of the framework is evaluated through case studies and expert interviews, demonstrating its potential to uncover complex patterns and facilitate a deeper understanding of multi-objective optimization solution sets. A demo website of ParetoLens is available at https://dva-lab.org/paretolens/.

Figures (8)

• Figure 1: The nested model for visualization design Munzner2009Munzner2014 is shown on the left side. The corresponding design stages regarding this work are illustrated on the right side.
• Figure 2: The overview of the framework with two main stages. After the required data about solutions and reference sets are exported from the computation framework, the data preprocessing stage converts the collected data into an intermediate transmission format in JSON. The file is then loaded into the visualization frontend to enable the interactive visual inspection and exploration of solutions.
• Figure 3: The interface of ParetoLens comprises a set of visualization components (A.1-A.3, B.1-B.2) aimed at visually inspecting and exploring solutions from two major analytical aspects: (A) decision and objective space analysis, and (B) trade-off and correlation analysis.
• Figure 6: The result of RVEA on the DTLZ3 problem in Case Study 1. (A) A snapshot of the main interface. In the objective space projection result, Alice has made several noteworthy observations regarding distributions, color mappings, and the reference density map. (B) A detailed description of the dominance relationship between two solutions (#63 and #44). They are considered mutual nearest neighbors and cannot dominate each other. (C) A detailed illustration of a pair of mutual nearest neighbors. They are closely positioned in the objective space while maintaining a considerable distance from each other in the decision space. (D) The PCP when a reference density map area shown in (A.3) is selected, with solution #64 highlighted.
• Figure 7: The corresponding selections of solutions in the distance histogram. (a) Group 1: The selected solutions are associated with areas where duplicated solutions exist within each reference density map area. (b) Group 2: The majority of solutions with average distances are included in this group.