Research map / Time series / Foundation models for time series
Time series foundation models
Adapt pretrained time-series models for zero-shot forecasting with in-context tuning and adapters.
71 accepted papers at ICML, NeurIPS, ICLR, CVPR and AAAI (2021–2026) · an approach in Foundation models for time series, keyword cluster “shot · series foundation · llm”
State History Tried Baselines 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.
6 papers in 2023–24 and 64 in 2025–26 (at ICML, NeurIPS, ICLR, CVPR and AAAI); its share of the topic grew, from 5.6% to 25.2%.
Open questions the data points to
- Papers report TimesNet on ETTm1 (mSE) anywhere from 0.172 to 0.383: comparisons against it depend on whose number you take.
- Papers report TimesNet on Electricity (mSE) anywhere from 0.14 to 0.193: comparisons against it depend on whose number you take.
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
- 2021 · Meta-Learning Framework with Applications to Zero-Shot Time-Series Forecasting (AAAI 2021 · 74 citations)
- 2023 · ForecastPFN: Synthetically-Trained Zero-Shot Forecasting (NeurIPS 2023 · 14 citations)
- 2024 · MOMENT: A Family of Open Time-series Foundation Models (ICML 2024 · 27 citations)
- 2025 · Federated Foundation Models on Heterogeneous Time Series (AAAI 2025 · 12 citations)
- 2026 · Kronos: A Foundation Model for the Language of Financial Markets (AAAI 2026 · 5 citations)
Earliest papers here
- Meta-Learning Framework with Applications to Zero-Shot Time-Series Forecasting (AAAI 2021 · 74 citations)
- ForecastPFN: Synthetically-Trained Zero-Shot Forecasting (NeurIPS 2023 · 14 citations)
- Improving *day-ahead* Solar Irradiance Time Series Forecasting by Leveraging Spatio-Temporal Context (NeurIPS 2023 · 9 citations)
What has been tried
30 papers here report 304 measured changes (a component added, removed or swapped, compared with a baseline): helped 68%, no clear effect 8%, hurt 19%, 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 |
|---|---|---|---|---|
| Initialize MOMENT-small randomly instead of with Flan-T5 weights | Time Series Pile pre-training · Training loss | Helped | MOMENT: A Family of Open Time-series Foundation Models (ICML 2024 · 27 citations) | |
| Set ATM top-k to 3 rather than 1 or 2 | Federated pretraining, L_m=16 · MSE | .432 -> .421; .427 -> .421 | Helped | Federated Foundation Models on Heterogeneous Time Series (AAAI 2025 · 12 citations) |
| Add convolutional filtering before patching | Zero-shot MTS benchmarks · MSE | 0.315 -> 0.653 (ECL) | Helped | TimePFN: Effective Multivariate Time Series Forecasting with Synthetic Data (AAAI 2025 · 11 citations) |
| add prompt adaption | ETTh1-336 · MSE | 0.495 -> 0.480 | Helped | Time-FFM: Towards LM-Empowered Federated Foundation Model for Time Series Forecasting (NeurIPS 2024 · 10 citations) |
| Add PAAS to TAFAS | ETTh1, H=720 · MSE | 0.740 -> 0.704 | Helped | Battling the Non-stationarity in Time Series Forecasting via Test-time Adaptation (AAAI 2025 · 9 citations) |
| pretrain with more UKB data | HCP-Aging age · Pearson correlation (ρ) | 0.659 (.043); 0.768 (.012); 0.813 (.015); 0.844 (.030) | Helped | Brain-JEPA: Brain Dynamics Foundation Model with Gradient Positioning and Spatiotemporal Masking (NeurIPS 2024 · 5 citations) |
| Replace gated fusion with ARM | Seven benchmark average · MSE | 0.1982 -> 0.1940 | Helped | TS-RAG: Retrieval-Augmented Generation based Time Series Foundation Models are Stronger Zero-Shot Forecaster (NeurIPS 2025 · 4 citations) |
| Use MAE Large instead of MAE Base | Long-term TSF average · MSE | 0.309 -> 0.311 | Hurt | VisionTS: Visual Masked Autoencoders Are Free-Lunch Zero-Shot Time Series Forecasters (ICML 2025 · 3 citations) |
| use TimeFlow Loss instead of diffusion loss | TSLib (six-dataset average) · MSE | 0.314 -> 0.290 | Helped | Sundial: A Family of Highly Capable Time Series Foundation Models (ICML 2025 · 2 citations) |
| Add causal patch-based scaling instead of naive global scaling | Observability validation data · Forecasting performance | Helped | This Time is Different: An Observability Perspective on Time Series Foundation Models (NeurIPS 2025 · 2 citations) | |
| Replace dual-branch energy-aware spectral masking with multi-band masking | TSLib, zero-shot average · MSE | 0.350 -> 0.399 | Hurt | SEMPO: Lightweight Foundation Models for Time Series Forecasting (NeurIPS 2025 · 2 citations) |
| Use Student-t mixture instead of Gaussian mixture | ID and OOD forecasting · NLL | Helped | Towards Neural Scaling Laws for Time Series Foundation Models (ICLR 2025 · 2 citations) |
Baseline numbers that disagree
The same baseline on the same benchmark and metric, reported more than 5% apart by different papers here. Training budgets, data splits and reimplementations differ, so state which number you compare with.
| Baseline | Benchmark | Metric | Reported | Papers |
|---|---|---|---|---|
| TimesNet | ETTm1 | mSE | 0.172 – 0.383 | 2 papers, e.g. ICML 2024 and NeurIPS 2024 |
| TimesNet | Electricity | mSE | 0.14 – 0.193 | 2 papers, e.g. ICML 2024 and NeurIPS 2024 |
| MOIRAI | Electricity zero-shot | MSE | 0.188 – 0.241 | 2 papers, e.g. ICML 2025 and ICML 2025 |
What reviewers will expect
Datasets and baselines named in the 37 papers of this approach from 2025–2026 that have a full-text summary. Leaving out the common ones is a frequent review question.
Datasets
- Weather · 16 of 37 papers
- ETTh1 · 15 of 37 papers
- ETTm1 · 14 of 37 papers
- ETTh2 · 14 of 37 papers
- ETTm2 · 14 of 37 papers
- Electricity · 10 of 37 papers
- Traffic · 8 of 37 papers
- Exchange · 5 of 37 papers
Baselines
- PatchTST · 17 of 37 papers
- Moirai · 11 of 37 papers
- iTransformer · 10 of 37 papers
- DLinear · 9 of 37 papers
- TimesFM · 9 of 37 papers
- TimesNet · 7 of 37 papers
- MOMENT · 7 of 37 papers
- Chronos · 7 of 37 papers
Recent preprints
Loading recent arXiv preprints…
Who publishes on it
Institutions on the most papers of this approach since 2023 (of 70 papers; from OpenAlex and the papers' affiliations).
- Salesforce · 4 papers
- Huawei · 4 papers
- Carnegie Mellon University · 3 papers
- Google · 3 papers
- Hong Kong University of Science and Technology · 3 papers
- National University of Singapore · 3 papers
- Nanyang Technological University · 3 papers
- The Hong Kong Polytechnic University · 2 papers
- Amazon · 2 papers
- Shanghai Jiao Tong University · 2 papers
Use it
Download BibTeX of 23 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
- LLM-based time series forecasting (90 papers)
- Time-series language reasoning (44 papers)
Elsewhere in the topic
- Time series representation learning (115 papers)
- Time series classification (61 papers)
- Satellite image time series (28 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).