MSC-CMA-ES: Structure-Aware Restarts for CMA-ES via Cyclic Nearest-Better Basin Discovery
Researchers have developed a new optimization strategy called MSC-CMA-ES, designed to improve the performance of the CMA-ES algorithm in multimodal search scenarios. This method introduces structure-aware restarts by partitioning search spaces into basins of attraction and seeding restarts within these basins. Evaluations on various benchmark suites indicate that MSC-CMA-ES excels on composition functions, demonstrating significantly higher coverage than other algorithms, though it shows a trade-off between landscape discovery and deep-target coverage on basic functions. AI