Reproduce “Scaling MAP-Elites to Deep Neuroevolution” on OpenResearch
Turn your favorite coding agents into research agents that review literature, develop hypotheses, and run experiments.
Universal app for Apple silicon and Intel. Requires macOS 11 or later.
About this paper
Cédric Colas, Joost Huizinga, Vashisht Madhavan, Jeff Clune
ME-ES combines a MAP-Elites behavioral repertoire with Evolution Strategies to optimize neural-network controllers. It retains the controller in each behavior niche while improving candidates through perturbations, targeting performance, novelty, or alternating between them. In simulated Ant damage tests, performance-directed variants recover in all 12 cases; in Ant Maze, alternating optimization escapes a trap that NSR-ES does not, though others escape.
Summary by alphaXiv.
Tip: change arxiv.org to autoarxiv.org in any arXiv abstract link to open the paper here.