Prépublication · 2026
One2N: Recursive refinement breaks the resolution barrier for Earth-system AI
Yanyan Tan, Xiangjun Tian et al. — Monde
Source : arXiv
Résumé
Earth-system AI models approach kilometre-scale resolution, but halving grid spacing quadruples the grid-point count and end-to-end training cost. We introduce One2N, a recursive incremental framework extending a coarse model through $N$ levels of cross-resolution residual correction. At each level the parent is frozen and excluded from backpropagation, so its activations and optimizer states are never retained; only a tier-specific module is trained. A single refinement level on the NeuralGCM-0.7$^\circ$ baseline yields the two-level One2N-0.07$^\circ$, trained on one 80 GB GPU. Against HRES-…