Meta-Black-Box Optimization for Evolutionary Algorithms: Review and Perspective

Abstract

Meta-Black-Box Optimization (MetaBBO) emerges as a pivotal solution to enhance or discover optimization algorithms automatically via meta-learning. This comprehensive survey integrates the extant research within MetaBBO for Evolutionary Algorithms (EAs) to develop a consistent community. Specifically, a mathematical model for MetaBBO is established, and the scopes are clarified. The optimization object in MetaBBO for EAs is explored, providing insights into potential configurations. MetaBBO methodologies are introduced next, reflecting the state-of-the-art from a meta-level perspective. Additionally, related benchmarks, evaluation metrics, and platforms are overviewed, helping for those engaged in MetaBBO for EA. Finally, an outlook on future research is concluded.

Keywords: Black-Box Optimization, Meta-Black-Box Optimization, Evolutionary Algorithms, automatic algorithm design, learn to optimize, comprehensive survey