Reproduce “Action Learning for Scientific Simulations” 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
Sarah Liaw, Yuanqi Du, Yilun Du
LEAP learns a reusable cost to score trajectories from paths. It contrasts paths with samples from a reference process and combines a path log-ratio with the reference action. On a 31-body orbital benchmark, its model reaches 0.228 NRMSE versus about 1 for local models; molecular tests reuse the action for new requests. These results apply to the tested settings.
Summary by alphaXiv.
Tip: change arxiv.org to autoarxiv.org in any arXiv abstract link to open the paper here.