Quality Indicators in Evolutionary Multiobjective Optimization

Since the early days of multiobjective evolutionary algorithms (MOEAs), comparing their performance has been essential. Initially, MOEAs were compared through visual examinations of their resulting Pareto front approximations (PFAs). However, the need for quantitative assessment soon became evident. In this regard, quality indicators (QIs) emerged as the primary tools to numerically analyze the performance of MOEAs. Although the most popular utilization of QIs is focused on assessing the final PFA generated by an MOEA, their use is not restricted to this aim. QIs can also be used to quantitatively understand the solutions in the decision space and/or the online behavior of MOEAs on unconstrained multiobjective optimization problems (MOPs) and their main variants, such as constrained MOPs, dynamic MOPs, MOPs under uncertainty, among others. Beyond analysis, QIs are integral to mechanisms that impact population survival and archiving of solutions. They also serve as feedback mechanisms for online multiobjective hyper-heuristics, as well as performance assessment tools during automated algorithm design and configuration. Furthermore, QIs are currently the focus of theoretical analysis to design efficient algorithms to solve or approximate the indicator-based subset selection problem, where mathematical properties such as non-decreasing monotonicity and submodularity are key factors. Hence, QIs are at the core of evolutionary multiobjective optimization (EMO), where many questions and problems are still open.

This special session aims to concentrate the efforts of the EMO community on the study of QIs. Consequently, we aim to explore the current advances in the utilization of QIs in EMO for performance assessment and algorithm design. Our goal is to delve into the challenges and opportunities of exploiting the power of QIs to understand MOEAs, proving cutting-edge theoretical and empirical results.

The main topics of this special session include but are not limited to the following ones:

• Proposal of new QIs
• Theoretical and empirical analysis of QIs
• Indicator-based subset selection & archiving
• Indicator-driven MOEA configuration & design
• Runtime quality indicators
• QIs for decision space
• QIs for interactive algorithms
• QIs for multiobjective hyper-heuristics
• Explainable QIs
• Other emerging topics on QIs