Research map / Time series / Causal and generative time series
Dynamical system reconstruction
Learn latent nonlinear or linear dynamical systems from time series data.
90 accepted papers at ICML, NeurIPS, ICLR, CVPR and AAAI (2016–2026) · an approach in Causal and generative time series, keyword cluster “conformal · causal discovery · predictions”
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.
23 papers in 2023–24 and 27 in 2025–26 (at ICML, NeurIPS, ICLR, CVPR and AAAI); its share of the topic shrank, from 14.8% to 11.8%.
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 · Efficient Multi-Instance Learning for Activity Recognition from Time Series Data Using an Auto-Regressive Hidden Markov Model (ICML 2016 · 39 citations)
- 2017 · Learning the Morphology of Brain Signals Using Alpha-Stable Convolutional Sparse Coding (NeurIPS 2017 · 6 citations)
- 2018 · Probabilistic Recurrent State-Space Models (ICML 2018 · 8 citations)
- 2019 · Learning Procedural Abstractions and Evaluating Discrete Latent Temporal Structure (ICLR 2019 · 9 citations)
- 2020 · HiPPO: Recurrent Memory with Optimal Polynomial Projections (NeurIPS 2020 · 148 citations)
- 2021 · Affect2MM: Affective Analysis of Multimedia Content Using Emotion Causality (CVPR 2021 · 38 citations)
- 2022 · Reconstructing Nonlinear Dynamical Systems from Multi-Modal Time Series (ICML 2022 · 9 citations)
- 2023 · Causal Recurrent Variational Autoencoder for Medical Time Series Generation (AAAI 2023 · 75 citations)
- 2024 · Learning the Causal Structure of Networked Dynamical Systems under Latent Nodes and Structured Noise (AAAI 2024 · 5 citations)
- 2025 · ERCI: An Explainable Experience Replay Approach with Causal Inference for Deep Reinforcement Learning (AAAI 2025 · 2 citations)
- 2026 · RamanBench: A Large-Scale Benchmark for Machine Learning on Raman Spectroscopy (NeurIPS 2026 · 1 citation)
Earliest papers here
- Efficient Multi-Instance Learning for Activity Recognition from Time Series Data Using an Auto-Regressive Hidden Markov Model (ICML 2016 · 39 citations)
- Interpretable Nonlinear Dynamic Modeling of Neural Trajectories (NeurIPS 2016 · 23 citations)
- The Segmented iHMM: A Simple, Efficient Hierarchical Infinite HMM (ICML 2016 · 7 citations)
What has been tried
66 papers here report 570 measured changes (a component added, removed or swapped, compared with a baseline): helped 66%, no clear effect 10%, hurt 20%, mixed 4%. Papers mostly publish changes that helped, so "helped" is over-represented.
Ablations in recent papers
| What was changed | Measured on | Result | Effect | Paper |
|---|---|---|---|---|
| Combine lag-moment features F with inverse lag-moment features T | Under the sufficient condition in Theorem 2 · identifiability gap | Helped | Learning the Causal Structure of Networked Dynamical Systems under Latent Nodes and Structured Noise (AAAI 2024 · 5 citations) | |
| Remove the DSR reconstruction loss | Multimodal dynamical-system reconstruction · reconstruction performance | Hurt | Integrating Multimodal Data for Joint Generative Modeling of Complex Dynamics (ICML 2024 · 5 citations) | |
| Add GRU refinement to autoregressive forecasting | Ferry FC · RMSE | 0.0777 -> 0.0608 | Helped | Sequential Modeling of Complex Marine Navigation: Case Study on a Passenger Vessel (Student Abstract) (AAAI 2024 · 3 citations) |
| Add static covariates containing neuron spatial positions to TiDE | ZAPBench · MAE | Hurt | ZAPBench: A Benchmark for Whole-Brain Activity Prediction in Zebrafish (ICLR 2025 · 2 citations) | |
| remove GTF | LEMON · D_stsp | 2.73 ± 1.09 -> 3.77 ± 1.22 | Hurt | A scalable generative model for dynamical system reconstruction from neuroimaging data (NeurIPS 2024 · 2 citations) |
| Use the full emergence objective instead of predictive mutual information alone | Bit-string dataset · emergent feature recovery | Helped | Learning diverse causally emergent representations from time series data (NeurIPS 2024 · 2 citations) | |
| add the regularizer | dysts, 5%, 25%, and 50% noise · Coefficient error | Helped | Modeling Latent Non-Linear Dynamical System over Time Series (AAAI 2025 · 1 citation) | |
| Use squared instead of ReLU state-matrix parameterization | UEA-MTSCA Avg · accuracy | 67.8 -> 67.1 | Hurt | Oscillatory State-Space Models (ICLR 2025 · 1 citation) |
| use hierarchical rather than ensemble shPLRNNs | Lorenz-63 · D_stsp | 1.82±0.16 -> 0.394±0.014 | Helped | Learning Interpretable Hierarchical Dynamical Systems Models from Time Series Data (ICLR 2025 · 1 citation) |
| Set the forecasting order to k=90 rather than k=45 | M4-Weekly · average sMAPE | 7.74 -> 6.28 | Helped | Towards Understanding Evolving Patterns in Sequential Data (NeurIPS 2024 · 1 citation) |
| use learnable frequency decomposition instead of a uniform frequency mask | ECL, Weather, and ETTh1 forecasting · MSE | ECL T=48: 0.137 vs 0.150; Weather T=48: 0.131 vs 0.134; ETTh1 T=48: 0.333 vs 0.329 | Mixed | SKOLR: Structured Koopman Operator Linear RNN for Time-Series Forecasting (ICML 2025 · 0 citations) |
| Use a decomposed kernel | Catechol single solvent · MSE | 0.011 -> 0.009 | Helped | The Catechol Benchmark: Time-series Solvent Selection Data for Few-shot Machine Learning (NeurIPS 2025 · 0 citations) |
What reviewers will expect
Datasets and baselines named in the 19 papers of this approach from 2025–2026 that have a full-text summary. Leaving out the common ones is a frequent review question.
Baselines
- LRU · 4 of 19 papers
- Mamba · 3 of 19 papers
- Log-NCDE · 3 of 19 papers
- NRDE · 2 of 19 papers
- NCDE · 2 of 19 papers
- S5 · 2 of 19 papers
- ROCKET · 2 of 19 papers
Recent preprints
Loading recent arXiv preprints…
Who publishes on it
Institutions on the most papers of this approach since 2023 (of 50 papers; from OpenAlex and the papers' affiliations).
- Ruprecht-Karls-Universität Heidelberg · 3 papers
- Heidelberg University · 3 papers
- University College London · 3 papers
- Pittsburgh · 2 papers
- Hebrew University of Jerusalem · 2 papers
- Brown University · 2 papers
- Central Institute of Mental Health · 2 papers
- Amazon · 2 papers
- Google · 2 papers
- Imperial College London · 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
- Probabilistic and conformal forecasting (121 papers)
- State-space and latent dynamics forecasting (94 papers)
- Robust forecasting and evaluation (91 papers)
- Generative temporal point processes (53 papers)
Elsewhere in the topic
- Irregular series and imputation (59 papers)
- Diffusion models for time series (55 papers)
- Neural differential equations (52 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).