Research map / Time series / Causal and generative time series
State-space and latent dynamics forecasting
Forecast via deep state-space, latent-state and dynamics-based models with approximation theory.
94 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.
18 papers in 2023–24 and 28 in 2025–26 (at ICML, NeurIPS, ICLR, CVPR and AAAI); its share of the topic held steady, from 11.6% to 12.2%.
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 · Predicting ICU Mortality Risk by Grouping Temporal Trends from a Multivariate Panel of Physiologic Measurements (AAAI 2016 · 66 citations)
- 2017 · Modeling the Intensity Function of Point Process Via Recurrent Neural Networks (AAAI 2017 · 206 citations)
- 2018 · Deep State Space Models for Time Series Forecasting (NeurIPS 2018 · 429 citations)
- 2019 · Time-series Generative Adversarial Networks (NeurIPS 2019 · 511 citations)
- 2020 · Benchmarking Deep Learning Interpretability in Time Series Predictions (NeurIPS 2020 · 93 citations)
- 2021 · Conformal Time-series Forecasting (NeurIPS 2021 · 33 citations)
- 2022 · Clustering Interval-Censored Time-Series for Disease Phenotyping (AAAI 2022 · 18 citations)
- 2023 · An Extreme-Adaptive Time Series Prediction Model Based on Probability-Enhanced LSTM Neural Networks (AAAI 2023 · 23 citations)
- 2024 · Market-GAN: Adding Control to Financial Market Data Generation with Semantic Context (AAAI 2024 · 11 citations)
- 2025 · Neural Conformal Control for Time Series Forecasting (AAAI 2025 · 5 citations)
- 2026 · Time Series Forecasting via Direct Per-Step Probability Distribution Modeling (AAAI 2026 · 2 citations)
Earliest papers here
- Predicting ICU Mortality Risk by Grouping Temporal Trends from a Multivariate Panel of Physiologic Measurements (AAAI 2016 · 66 citations)
- An Efficient Time Series Subsequence Pattern Mining and Prediction Framework with an Application to Respiratory Motion Prediction (AAAI 2016 · 5 citations)
- Convolution Kernels for Discriminative Learning from Streaming Text (AAAI 2016 · 5 citations)
What has been tried
74 papers here report 673 measured changes (a component added, removed or swapped, compared with a baseline): helped 67%, no clear effect 10%, hurt 18%, 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 |
|---|---|---|---|---|
| Use C-TimesBlock instead of RNN | DJI stocks, context alignment · L_d | 0.626 -> 0.023 | Helped | Market-GAN: Adding Control to Financial Market Data Generation with Semantic Context (AAAI 2024 · 11 citations) |
| add coverage loss (with efficiency loss) | five datasets, one-step predictions · CS | 0.154 -> 0.104 (flu); 0.0883 -> 0.0551 (covid-19); 0.0466 -> 0.0221 (weather); 0.115 -> 0.0289 (smd); 0.0472 -> 0.0161 ( | Helped | Neural Conformal Control for Time Series Forecasting (AAAI 2025 · 5 citations) |
| Use a well-specified velocity-field kernel to estimate the synthetic deformation parameters | Synthetic time-series graph deformation · average relative error | 0.10, 0.03, 0.17 | Helped | Shape analysis for time series (NeurIPS 2024 · 1 citation) |
| Replace the linear base predictor with an MLP | ECL · MSE | 0.1646 ± 0.0001 -> 0.1616 ± 0.0005 | Helped | LETS Forecast: Learning Embedology for Time Series Forecasting (ICML 2025 · 0 citations) |
| Use λ=1 rather than λ>1 subsequences per sequence | privacy accounting after 100 training steps · δ(ε) | Helped | Privacy Amplification by Structured Subsampling for Deep Differentially Private Time Series Forecasting (ICML 2025 · 0 citations) | |
| Replace independent random point masking with segment-based masking | Average and zero substitution · CPD | 0.175 -> 0.177; 0.350 -> 0.366 | Helped | TIMING: Temporality-Aware Integrated Gradients for Time Series Explanation (ICML 2025 · 0 citations) |
| Remove all normalization layers | triangle wave and pulse patterns, H=96 · performance | Helped | SynTSBench: Rethinking Temporal Pattern Learning in Deep Learning Models for Time Series (NeurIPS 2025 · 0 citations) | |
| Add learned guidance to the transformation rather than mapping every sample to a single fixed manifold | IEEE SPC22 · RMSE | 19.18 -> 16.25 | Helped | Shifting the Paradigm: A Diffeomorphism Between Time Series Data Manifolds for Achieving Shift-Invariancy in Deep Learning (ICLR 2025 · 0 citations) |
| Replace ordinary convolutions with strided convolutions in predictive inference | DNS speech separation, prediction length 3 · SI-SNRi (dB) | 4.61 -> 5.47 | Helped | SOI: Scaling Down Computational Complexity by Estimating Partial States of the Model (NeurIPS 2024 · 0 citations) |
| Add adapt-STE rather than using STE | ECG · AUPRC | 0.6152±0.0007 -> 0.7183±0.0019 | Helped | Optimal Information Retention for Time-Series Explanations (ICML 2025) |
What reviewers will expect
Datasets and baselines named in the 20 papers of this approach from 2025–2026 that have a full-text summary. Leaving out the common ones is a frequent review question.
Datasets
- Electricity · 2 of 20 papers
- Traffic · 2 of 20 papers
- PAM · 2 of 20 papers
- Boiler · 2 of 20 papers
- Epilepsy · 2 of 20 papers
Baselines
- iTransformer · 3 of 20 papers
- PatchTST · 3 of 20 papers
- DLinear · 3 of 20 papers
- TimeMixer · 2 of 20 papers
- Naive · 2 of 20 papers
- WinIT · 2 of 20 papers
- Dynamask · 2 of 20 papers
- TimeX · 2 of 20 papers
Recent preprints
Loading recent arXiv preprints…
Who publishes on it
Institutions on the most papers of this approach since 2023 (of 46 papers; from OpenAlex and the papers' affiliations).
- Tsinghua University · 3 papers
- Microsoft · 3 papers
- Amazon · 3 papers
- Yale University · 2 papers
- University of California, San Diego · 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)
- Dynamical system reconstruction (90 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).