AI Paper Atlas

Research map / Probabilistic and scientific modeling / Molecules and proteins

Physically-Guided Data-Space Rectified Flow for Precipitation Nowcasting

Wenjie Luo, chaorong li, Chuanhu Deng, Zhuo Wang

ICML 2026 · Probabilistic and scientific modeling › Molecules and proteins · approach: Flow matching generative modeling

Read the paperCheck your idea against 137,217 papers

AI-extracted. The measured effects below were read by AI from the paper's full text (no summary yet). Numbers and claims can be wrong or missing context, so check the paper before you cite or build on it.

Measured effects reported in the paper

What was changedWhereMeasuredEffect
add conditional rectified flow training to backbone (ablation)U-KAN backbone · MeteoNetCSI-M: 0.3793 -> 0.4021 (+6.01%)Improved
add conditional rectified flow (v-pred) to backbone (ablation)U-KAN backbone · MeteoNetSSIM: 0.8059 -> 0.6812 (-15.47%)Hurt
switch from v-pred to x-pred (data-space prediction) (ablation)PDRF · MeteoNetSSIM: 0.6956 -> 0.8348 (+19.99%)Improved
switch from v-pred to x-pred (data-space prediction) (ablation)PDRF · CIKMSSIM: 0.5561 -> 0.6576 (+18.25%)Improved
switch from v-pred to x-pred (data-space prediction) (ablation)PDRF · SEVIRCSI-M: 0.3552 -> 0.3522 (-0.84%)Hurt
add Semi-Lagrangian physics regularizer L_phys (ablation)PDRF · SEVIRCSI-M: 0.3522 -> 0.3622 (+2.84%)Improved
add Semi-Lagrangian physics regularizer L_phys (ablation)PDRF · MeteoNetCSI-M: 0.4289 -> 0.4386Improved
increase physical weight lambda to 0.5 or more (ablation)PDRF · MeteoNetCSI-M: 0.4386 -> 0.4235 (lambda=0.5), 0.4129 (lambda=1.0)Hurt
use x-pred instead of v-pred at lambda=0.1 (ablation)PDRF · MeteoNetCSI-M: 0.4115 -> 0.4386Improved
compute loss in velocity domain instead of x-space (ablation)PDRF · MeteoNetCSI-M: 0.4281 -> 0.4386Improved
add CGSTF to backbone with CRF (ablation)PDRF · SEVIRCSI-M: 0.3456 -> 0.3389 (-1.94%)Hurt
add FCM (ablation)PDRF · CIKMCSI-40: 0.1434 -> 0.1567 (+9.28%)Improved

Closest papers by meaning

Writing a paper in this area? scime draft is a free LaTeX editor with AI included, real-time co-authors and a check of your references and venue rules before you submit. It also imports OpenAI Prism projects.

Open scime draft

Get the next scime Dispatch and a monthly field update

Free. We send a confirmation link first; every email has a one-click unsubscribe. Privacy