MAP-Elites

Multi-dimensional Archive of Phenotypic Elites (MAP-Elites) is a Quality-Diversity (QD) algorithm: it illuminates a search space by building an archive of the best-performing solution for every defined behavioral characteristic.

MAP-Elites is the flagship Quality-Diversity (QD) search algorithm, co-developed by Mouret and Clune in 2015. It partitions a high-dimensional solution space into a multi-dimensional grid, or feature map, based on user-defined behavioral characteristics (BCs). The core function is to fill each cell of this grid with the highest-performing 'elite' solution found for that specific BC niche. This process, unlike single-objective optimization, systematically generates a diverse repertoire of high-quality solutions. The technique has proven invaluable in fields like evolutionary robotics, creative AI, and neural architecture search, providing not just one answer, but a comprehensive map of optimal trade-offs across the problem space.

https://www.loria.fr/~jbmouret/qd.html