SDECast: Neural SDEs bring continuous-time probabilistic weather forecasting with 5-day skill
scott_linderman · x · 2026-10-14
- SDECast (accepted at NeurIPS 2026 AI for Stochastic Dynamics & Sim2Science workshops) uses Neural SDEs for continuous-time probabilistic weather forecasting.
- Existing ML weather models rely on autoregressive rollouts at fixed resolution, accumulating errors at short steps; SDECast extends SDE Matching to learn stochastic dynamics directly in physical space without repeated SDE simulation during training.
- It recovers meaningful drift dynamics on simulated geophysical flow and produces skillful probabilistic forecasts up to 5 days ahead at hourly global resolution.
- Paper and code are public; the authors also present Gauge Matching, another SDE Matching extension.
More from Research
- SURGE Surgical Foundation Model: Perception Saturates After Just One Labeled Video — maier_ak · 2026-10-14
- Prompteina: First Text-Conditioned All-Atom Protein Structure Generation Model — PangWeiKoh · 2026-10-14
- Correction Training frees diffusion models from CFG as looped architectures gain traction — CSProfKGD · 2026-10-14
- FramePack-U: Training-free constant-memory frame packing boosts long-video VLM benchmarks by up to 12.9% — UNC-ChapelHill · 2026-10-14
- CineRecon Benchmark: 12 Frontier Models Struggle to Reconstruct Editable 3D Scenes from Video — hkuhk · 2026-10-14
- Meituan's ClawLoop: Self-Improving Agents via Teacher-Free Harness-Native Training Loops — meituan-longcat · 2026-10-14