Reproduce “GraphCast-Mamba: Learning Non-Markovian Corrections for Weather Forecasting” 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
Ilya Vilkoviskiy, Liang-Hong Mo, Suozhi Huang, Dmitry A. Abanin
GraphCast–Mamba adds a memory-bearing correction branch to frozen GraphCast, which predicts each six-hour step from two recent atmospheric states. The branch carries state across forecast steps and adjusts the backbone’s predictions; it is trained on ERA5 data. On 122 ten-day forecasts, GraphCast loss falls 14.42% overall and 21.66% at day 10, while scope remains one 1° backbone and training realization.
Summary by alphaXiv.
Tip: change arxiv.org to autoarxiv.org in any arXiv abstract link to open the paper here.