Automatically Designing State-of-the-Art Multi- and Many-Objective Evolutionary Algorithms.

Area of Science:

• Computational intelligence
• Optimization algorithms
• Evolutionary computation

Background:

• Existing multiobjective evolutionary algorithms (MOEAs) require extensive parameter tuning.
• Identifying state-of-the-art MOEAs necessitates considering automatic configuration, diverse setups, and multiple performance metrics.

Purpose of the Study:

• To automatically devise MOEAs with verified state-of-the-art performance for multi- and many-objective continuous optimization.
• To extend a configurable algorithmic framework for automated MOEA design.
• To propose a multiobjective formulation for automatic MOEA design, addressing metric disagreement in many-objective optimization.

Main Methods:

• Extending a configurable algorithmic framework with more MOEAs, evolutionary algorithms, and search paradigms.
• Employing an automatic configuration method to instantiate high-performing MOEA designs.
• Developing a multiobjective formulation for automatic MOEA design to optimize multiple performance metrics.

Main Results:

• Automatically designed MOEAs demonstrate state-of-the-art performance in continuous optimization.
• The extended framework facilitates the automated creation of effective MOEA designs.
• The multiobjective formulation for design ensures robust performance across various metrics, especially in many-objective scenarios.

Conclusions:

• Automated design of MOEAs is a viable approach to achieve state-of-the-art performance.
• A configurable framework and multiobjective optimization for design are crucial for developing advanced MOEAs.
• The proposed methods offer a systematic way to discover high-performing MOEAs for complex optimization problems.