AI Paper Atlas

Research map / Image generation / Fast diffusion sampling

GenDA: Generative Data Assimilation on Complex Urban Areas via Classifier-Free Diffusion Guidance

Francisco Giral, Álvaro Manzano Sevillano, Ignacio Gomez Perez, Ricardo Vinuesa, Soledad Le Clainche

ICML 2026 · Image generation › Fast diffusion sampling · approach: Flow matching distillation · 0 citations (OpenAlex)

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
use classifier-free guidance with sensor-conditioned and unconditional branches instead of direct observation conditioning (same backbone)GenDA multiscale graph diffusion · Urban wind reconstruction, 8000 obsRRMSE: 0.189 -> 0.136Improved
add the multiscale reduced-mesh hierarchy (ablation)GenDA · Urban wind reconstruction, 300 obsRRMSE: 0.399 -> 0.312Improved
add the multiscale reduced-mesh hierarchy (ablation)GenDA · Urban wind reconstruction, 3000 obsRRMSE: 0.254 -> 0.192Improved
add the multiscale reduced-mesh hierarchy (ablation)GenDA · Urban wind reconstruction, 8000 obsRRMSENo clear effect
add multiscale message passing to the supervised MeshGraphNet baselineMGN · Urban wind reconstruction, 300 obsRRMSE: 0.426 -> 0.412No clear effect
raise guidance weight from 1.0 to 2.0 (ablation)GenDA · Urban wind reconstruction, 300 obsRRMSE: 0.311 -> 0.298Improved
raise guidance weight from 2.0 to 3.0 (ablation)GenDA · Urban wind reconstruction, 8000 obsRRMSE: 0.145 -> 0.481Hurt
raise guidance weight from 2.0 to 3.0 (ablation)GenDA · Urban wind reconstruction, 3000 obsRRMSE: 0.183 -> 0.291Hurt
use guidance weight 1 (purely conditional) instead of guided sampling (ablation)GenDA · Urban wind reconstruction (various obs)RRMSEHurt
switch from random to cloud-based sensor placement at test timeGenDA · Sensor sampling strategy (100-4000 obs)RRMSE: 0.291 -> 0.354Hurt
switch from random to trajectory-based sensor placement at test timeGenDA · Sensor sampling strategy (100-4000 obs)RRMSE: 0.291 -> 0.345Hurt
increase number of assimilated observationsGenDA ensembles · Held-out sensor CRPSCRPS: 0.058 -> 0.040Improved

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