Reproduce “Skillful and realistic precipitation forecasting with reinforced diffusion” 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
Ozan Sener, Anand Avati, David Hafner, Jierui Lin, Laura Manduchi, Xiaoying Pang, Langwen Huang, Torsten Hoefler and 2 more
FitzRoy forecasts hourly precipitation across the continental United States. It uses discrete diffusion to generate coherent rain fields, then reinforcement learning (RL) to favor samples scoring better against observations. On held-out data, it performs competitively with tested baselines through 24 hours, though gains over HREF are more modest; a Texas case study reports an extreme-rain signal up to 19 hours ahead.
Summary by alphaXiv.
Tip: change arxiv.org to autoarxiv.org in any arXiv abstract link to open the paper here.