Research map / Probabilistic and scientific modeling / Dynamical systems
Latent dynamics models
Learns continuous-time or identifiable latent states from irregular or high-dimensional observations.
111 accepted papers at ICML, NeurIPS, ICLR, CVPR and AAAI (2016–2026) · an approach in Dynamical systems, keyword cluster “physical · physics · simulation”
State History Tried 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.
27 papers in 2023–24 and 32 in 2025–26 (at ICML, NeurIPS, ICLR, CVPR and AAAI); its share of the topic shrank, from 12.0% to 8.3%.
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 · Interaction Networks for Learning about Objects, Relations and Physics (NeurIPS 2016 · 557 citations)
- 2017 · Correlated Cascades: Compete or Cooperate (AAAI 2017 · 45 citations)
- 2018 · Deep Generative Markov State Models (NeurIPS 2018 · 41 citations)
- 2019 · Attentive State-Space Modeling of Disease Progression (NeurIPS 2019 · 52 citations)
- 2020 · Dynamic Neural Relational Inference (CVPR 2020 · 48 citations)
- 2021 · Inductive Representation Learning in Temporal Networks via Causal Anonymous Walks (ICLR 2021 · 62 citations)
- 2022 · The least-control principle for local learning at equilibrium (NeurIPS 2022 · 15 citations)
- 2023 · EINNs: Epidemiologically-Informed Neural Networks (AAAI 2023 · 31 citations)
- 2024 · DySLIM: Dynamics Stable Learning by Invariant Measure for Chaotic Systems (ICML 2024 · 4 citations)
- 2025 · Uncover Governing Law of Pathology Propagation Mechanism Through A Mean-Field Game (NeurIPS 2025 · 0 citations)
- 2026 · Modeling Trend Dynamics with Variational Neural ODEs for Information Popularity Prediction (AAAI 2026 · 1 citation)
Earliest papers here
- Interaction Networks for Learning about Objects, Relations and Physics (NeurIPS 2016 · 557 citations)
- Learning and Forecasting Opinion Dynamics in Social Networks (NeurIPS 2016 · 72 citations)
- Using Social Dynamics to Make Individual Predictions: Variational Inference with a Stochastic Kinetic Model (NeurIPS 2016)
What has been tried
78 papers here report 759 measured changes (a component added, removed or swapped, compared with a baseline): helped 64%, no clear effect 9%, hurt 22%, mixed 5%. Papers mostly publish changes that helped, so "helped" is over-represented.
Ablations in recent papers
| What was changed | Measured on | Result | Effect | Paper |
|---|---|---|---|---|
| Add DySLIM regularization to 1-step | Kolmogorov Flow · MELR | 2.77 -> 1.84 | Helped | DySLIM: Dynamics Stable Learning by Invariant Measure for Chaotic Systems (ICML 2024 · 4 citations) |
| Replace the KKL observer with GRU architectures for the partially OVS system | Partially OVS systems · accuracy | Hurt | Learning Hybrid Dynamics Models with Simulator-Informed Latent States (AAAI 2024 · 2 citations) | |
| remove L_integ | 2-Moons · error rate (%) | 2.8 ± 0.7 -> 27.4 ± 20.3 | Hurt | Continuous Temporal Domain Generalization (NeurIPS 2024 · 1 citation) |
| Replace region-aware edge weights with an unweighted graph | Navier-Stokes · Prediction performance | Hurt | Sparse Diffusion Autoencoder for Test-time Adapting Prediction of Complex Systems (NeurIPS 2025 · 0 citations) | |
| Remove the translational expert | Glycolytic system · MSE | MSE decreases to 0.001 | Helped | Latent Mixture of Symmetries for Sample-Efficient Dynamic Learning (NeurIPS 2025 · 0 citations) |
| Add the continuity regularizer to the standard autoencoder | Circular motion · latent-state continuity | Helped | Continuity-Preserving Convolutional Autoencoders for Learning Continuous Latent Dynamical Models from Images (ICLR 2025 · 0 citations) | |
| Vary the low-rank adaptation rank | Motor Cortex Recordings · behavior reconstruction and prediction r² | No clear effect | Meta-Dynamical State Space Models for Integrative Neural Data Analysis (ICLR 2025 · 0 citations) | |
| Add linear feedback | 12 learning-residual levels · prediction error | Helped | Feedback Favors the Generalization of Neural ODEs (ICLR 2025 · 0 citations) | |
| Add regularization penalizing update deviations from geometric constraints | MNIST optimizees with Leaky ReLU or batch normalization · optimization speed | Hurt | Investigation into Training Dynamics of Learned Optimizers (Student Abstract) (AAAI 2024 · 0 citations) | |
| Increase the data signal for a concept by widening its visual separation | Synthetic color and size concepts · learning speed | 80% accuracy | Helped | Emergence of Hidden Capabilities: Exploring Learning Dynamics in Concept Space (NeurIPS 2024 · 0 citations) |
| Add PointNet++ representations to PH representations | dorsogna-1k · VE | 0.851±0.008 -> 0.931±0.004 | Helped | Neural Persistence Dynamics (NeurIPS 2024 · 0 citations) |
| Increase state dimension from n=2m to n=3m | Single pendulum · Log Likelihood | 7.292 ± 0.173 -> 7.445 ± 0.165 | Helped | Gated Inference Network: Inference and Learning State-Space Models (NeurIPS 2024 · 0 citations) |
Recent preprints
Loading recent arXiv preprints…
Who publishes on it
Institutions on the most papers of this approach since 2023 (of 59 papers; from OpenAlex and the papers' affiliations).
- Emory University · 2 papers
- Tsinghua University · 2 papers
- Tencent · 2 papers
- Imperial College London · 2 papers
- Google · 2 papers
- Cornell University · 2 papers
- Qiyuan Lab · 2 papers
- Beijing Institute of Technology · 2 papers
- Academy of Military Sciences · 2 papers
Use it
Download BibTeX of 40 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-space control and planning (78 papers)
- Data assimilation from sparse observations (58 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).