Neuroevolution

Neuroevolution studies how population-based search can create or adapt neural networks, including when gradients are unavailable or network structure itself must be discovered. Methods range from genetic algorithms and evolution strategies to NEAT and HyperNEAT, quality-diversity search, co-evolution, online plasticity, and language-model-guided program evolution; they often combine evolutionary search with reinforcement or supervised learning. Applications include robot and game control, spiking and recurrent models, compact edge systems, and scientific prediction, with evaluations spanning control tasks, image and sequence datasets, and partial differential equations. A central challenge is making costly, noisy fitness evaluations scale and transfer across settings while preserving robustness and behavioral diversity.