Research map / Probabilistic and scientific modeling / Dynamical systems
Data assimilation from sparse observations
Reconstructs physical dynamics from sparse measurements using generative assimilation.
58 accepted papers at ICML, NeurIPS, ICLR, CVPR and AAAI (2016–2026) · an approach in Dynamical systems, keyword cluster “physical · physics · simulation”
State History Tried Checklist Preprints Who Export
Where it stands
Not enough benchmark data Fewer than four papers report the same benchmark, so there is no record to follow.
8 papers in 2023–24 and 41 in 2025–26 (at ICML, NeurIPS, ICLR, CVPR and AAAI); its share of the topic grew, from 3.6% to 10.7%.
How this is decided
A benchmark is tracked when at least four papers of this approach report their own method on the same dataset and metric in at least two different years. Its record is the best number reported up to 2024 compared with the best reported in 2025–26: still improving means at least 1% better, flat means at most 0.2% better with at least two papers from 2025–26 trying. Numbers on a different scale from the rest (a fraction among percentages) and far outliers are left out. Results come from different papers with different setups, so they are signals, not head-to-head comparisons. The share of papers is counted at ICML, NeurIPS, ICLR, CVPR and AAAI.
How it got here
Most cited paper of each year
- 2016 · Variable Aperture Light Field Photography: Overcoming the Diffraction-Limited Spatio-Angular Resolution Tradeoff (CVPR 2016 · 14 citations)
- 2019 · Seeing the Wind: Visual Wind Speed Prediction with a Coupled Convolutional and Recurrent Neural Network (NeurIPS 2019 · 18 citations)
- 2021 · Variational Data Assimilation with a Learned Inverse Observation Operator (ICML 2021 · 19 citations)
- 2022 · Conditional Local Convolution for Spatio-Temporal Meteorological Forecasting (AAAI 2022 · 92 citations)
- 2023 · DYffusion: A Dynamics-informed Diffusion Model for Spatiotemporal Forecasting (NeurIPS 2023 · 21 citations)
- 2024 · Earthfarsser: Versatile Spatio-Temporal Dynamical Systems Modeling in One Model (AAAI 2024 · 47 citations)
- 2025 · SIDE: Socially Informed Drought Estimation Toward Understanding Societal Impact Dynamics of Environmental Crisis (AAAI 2025 · 2 citations)
- 2026 · MoCast: Learning Turbulent Motions Under Physical Guidance for Precipitation Nowcasting (AAAI 2026 · 2 citations)
Earliest papers here
- Variable Aperture Light Field Photography: Overcoming the Diffraction-Limited Spatio-Angular Resolution Tradeoff (CVPR 2016 · 14 citations)
- Spectral Learning of Dynamic Systems from Nonequilibrium Data (NeurIPS 2016 · 2 citations)
- Seeing the Wind: Visual Wind Speed Prediction with a Coupled Convolutional and Recurrent Neural Network (NeurIPS 2019 · 18 citations)
What has been tried
28 papers here report 213 measured changes (a component added, removed or swapped, compared with a baseline): helped 73%, no clear effect 8%, hurt 14%, mixed 6%. Papers mostly publish changes that helped, so "helped" is over-represented.
Ablations in recent papers
| What was changed | Measured on | Result | Effect | Paper |
|---|---|---|---|---|
| add the local convolutional component | TaxiBJ+ · MAE | 2.8 -> 2.1 | Helped | Earthfarsser: Versatile Spatio-Temporal Dynamical Systems Modeling in One Model (AAAI 2024 · 47 citations) |
| add social input | California drought severity estimation · MAE | 35.28 -> 30.07 | Helped | SIDE: Socially Informed Drought Estimation Toward Understanding Societal Impact Dynamics of Environmental Crisis (AAAI 2025 · 2 citations) |
| use MPDPS instead of DPS | Supernova Explosion, observation setting 1 · VRMSE | 0.339 ± 0.116 -> 0.283 ± 0.026 (ρ=1%); 0.291 ± 0.033 -> 0.272 ± 0.025 (ρ=3%) | Helped | Generating Full-field Evolution of Physical Dynamics from Irregular Sparse Observations (NeurIPS 2025 · 2 citations) |
| Use fewer connection-Laplacian eigenvectors rather than channel-wise RBF positional encoding | High-resolution surface, α=1.5% · vector alignment | Helped | Implicit Gaussian process representation of vector fields over arbitrary latent manifolds (ICLR 2024 · 1 citation) | |
| Integrate partial physics in the Phy-SSM unit | Drone extrapolation · MAE | 8.426±1.351 -> 3.008±0.053 | Helped | A Generalizable Physics-Enhanced State Space Model for Long-Term Dynamics Forecasting in Complex Environments (ICML 2025 · 0 citations) |
| Increase the number of FINOLA paths from one to four (4×1024) | ImageNet-1K validation, 8×8–256×256 feature maps · PSNR | 25.4 -> 25.9; 25.8 -> 26.2; 25.8 -> 26.2; 25.7 -> 26.1; 25.3 -> 25.4; 24.6 -> 24.8 | Helped | Exploring Invariance in Images through One-way Wave Equations (ICML 2025 · 0 citations) |
| Add directional information to message passing | NACA 4-digit, 5% Uniform and 10% Uniform · MSE (×10^-4) | 75.124 -> 59.308; 59.064 -> 43.647 | Helped | Flow Field Reconstruction with Sensor Placement Policy Learning (NeurIPS 2025 · 0 citations) |
| Replace unbiased Monte Carlo estimation with Jensen-based biased estimation | Lorenz-63 · RMSE | 0.202 -> 0.363 | Hurt | FlowDAS: A Stochastic Interpolant-based Framework for Data Assimilation (NeurIPS 2025 · 0 citations) |
| Add retained flow tokens | PEMS04F short- and long-term forecasting · MAE | 0.08; 1.32 | Helped | FlowNet: Modeling Dynamic Spatio-Temporal Systems via Flow Propagation (NeurIPS 2025 · 0 citations) |
| Use DiT-style conditioning instead of FiLM | Gyrokinetic prediction · performance | No clear effect | GyroSwin: 5D Surrogates for Gyrokinetic Plasma Turbulence Simulations (NeurIPS 2025 · 0 citations) | |
| use residual prediction instead of direct state prediction | Kolmogorov flow, validation loss and 200-step rollouts · MSE accuracy | Helped | Thermalizer: Stable autoregressive neural emulation of spatiotemporal chaos (ICML 2025) | |
| Add physics-informed regularization | Simulation response recovery, 20 dB, 8×8 sampling, 2 targets · NRMSE | 1.0504 -> 0.0495 | Helped | NEAR: Neural Electromagnetic Array Response (ICML 2025) |
What reviewers will expect
Datasets and baselines named in the 23 papers of this approach from 2025–2026 that have a full-text summary. Leaving out the common ones is a frequent review question.
Baselines
- iTransformer · 3 of 23 papers
- STAEformer · 2 of 23 papers
- Senseiver · 2 of 23 papers
- LRTFR · 2 of 23 papers
Recent preprints
Loading recent arXiv preprints…
Who publishes on it
Institutions on the most papers of this approach since 2023 (of 49 papers; from OpenAlex and the papers' affiliations).
- Hong Kong University of Science and Technology · 3 papers
- The University of Queensland · 2 papers
- The University of Tokyo · 2 papers
- University of Illinois, Urbana Champaign · 2 papers
- Carnegie Mellon University · 2 papers
- National University of Singapore · 2 papers
- Microsoft · 2 papers
- University of North Carolina at Chapel Hill · 2 papers
- University of California, San Diego · 2 papers
- Zhejiang University · 2 papers
Use it
Download BibTeX of 32 key papersOpen the topic in the interactive map
Get alerts. Describe your idea in scime atlas and save it (free scime account): you get a weekly email when new preprints land close to it.
Related approaches
Same keyword cluster
- Physics-informed dynamics learning (127 papers)
- Latent dynamics models (111 papers)
- Latent-space control and planning (78 papers)
- Stochastic population dynamics (53 papers)
Elsewhere in the topic
- State-space inference for stochastic systems (104 papers)
- Switching and law-discovery dynamics (81 papers)
- Synaptic plasticity learning rules (72 papers)
- Latent models of neural activity (70 papers)
- Recurrent circuit models of inference (70 papers)
- Low-rank and diffusion neural dynamics (53 papers)
- Functional connectivity modeling (50 papers)
- Stable contractive dynamics (42 papers)
Benchmark numbers, ablations and baseline numbers are extracted by AI from each paper's full text and can be wrong; the paper is the authority. Data: results index of 2026-11-02, citation counts from OpenAlex (2026-10-09).