Reproduce “AlphaStar: An Evolutionary Computation Perspective” 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

Kai Arulkumaran, Antoine Cully, Julian Togelius

The authors interpret AlphaStar’s population-based training in StarCraft II through Lamarckian evolution, competitive co-evolution, and quality diversity. Agents learn through reinforcement learning and self-play, while population training copies weights and varies hyperparameters across agents facing varied opponents. AlphaStar beat professional player MaNa five games to none; this perspective offers an interpretation, not evidence these mechanisms caused the win.

Summary by alphaXiv.

Tip: change arxiv.org to autoarxiv.org in any arXiv abstract link to open the paper here.