Research map / Time series / Foundation models for time series
Time-series language reasoning
Combine time series and text for clinical reasoning and multimodal embedding.
44 accepted papers at ICML, NeurIPS, ICLR, CVPR and AAAI (2024–2026) · an approach in Foundation models for time series, keyword cluster “shot · series foundation · llm”
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.
5 papers in 2023–24 and 39 in 2025–26 (at ICML, NeurIPS, ICLR, CVPR and AAAI); its share of the topic grew, from 4.7% to 15.4%.
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
- 2024 · Time-MMD: Multi-Domain Multimodal Dataset for Time Series Analysis (NeurIPS 2024 · 15 citations)
- 2025 · ChatTime: A Unified Multimodal Time Series Foundation Model Bridging Numerical and Textual Data (AAAI 2025 · 28 citations)
- 2026 · MTP: Exploring Multimodal Urban Traffic Profiling with Modality Augmentation and Spectrum Fusion (AAAI 2026 · 5 citations)
Earliest papers here
- Time-MMD: Multi-Domain Multimodal Dataset for Time Series Analysis (NeurIPS 2024 · 15 citations)
- UniMTS: Unified Pre-training for Motion Time Series (NeurIPS 2024 · 5 citations)
- JoLT: Jointly Learned Representations of Language and Time-Series for Clinical Time-Series Interpretation (Student Abstract) (AAAI 2024 · 2 citations)
What has been tried
15 papers here report 142 measured changes (a component added, removed or swapped, compared with a baseline): helped 68%, no clear effect 13%, hurt 18%, mixed 1%. Papers mostly publish changes that helped, so "helped" is over-represented.
Ablations in recent papers
| What was changed | Measured on | Result | Effect | Paper |
|---|---|---|---|---|
| Remove autoregressive continuous pre-training | ZSTSF · forecasting performance | Hurt | ChatTime: A Unified Multimodal Time Series Foundation Model Bridging Numerical and Textual Data (AAAI 2025 · 28 citations) | |
| Vary the LLM backbone | Health · MSE | No clear effect | Time-MMD: Multi-Domain Multimodal Dataset for Time Series Analysis (NeurIPS 2024 · 15 citations) | |
| add rotation-invariant augmentation | 18 motion time-series datasets, zero-shot · accuracy | 27.1 -> 43.5 | Helped | UniMTS: Unified Pre-training for Motion Time Series (NeurIPS 2024 · 5 citations) |
| Generate pseudo labels for unlabeled molecules with a classifier in a semi-supervised approach | Molecules · controllability MAE | 54.65\% worse | Hurt | Text2Data: Low-Resource Data Generation with Textual Control (AAAI 2025 · 2 citations) |
| Add ECG images to time-series input | PTB-XL Super · AUC | 81.2 -> 82.7 | Helped | GEM: Empowering MLLM for Grounded ECG Understanding with Time Series and Images (NeurIPS 2025 · 2 citations) |
| add geometric pre-alignment to fMRI representations | ADNI · accuracy | 61.35 -> 61.62 | Helped | Brain Harmony: A Multimodal Foundation Model Unifying Morphology and Function into 1D Tokens (NeurIPS 2025 · 2 citations) |
| Increase code dimension from 8 to 32 | UEA 29 datasets · mean accuracy | 0.723 -> 0.721 | Hurt | Abstracted Shapes as Tokens - A Generalizable and Interpretable Model for Time-series Classification (NeurIPS 2024 · 2 citations) |
| add ITA only | EngineMT-QA component ablation · Average accuracy | 49.82 -> 50.94 | Helped | ITFormer: Bridging Time Series and Natural Language for Multi-Modal QA with Large-Scale Multitask Dataset (ICML 2025 · 0 citations) |
| add symbolic transformation | WESAD, full modalities · accuracy | 0.69 -> 0.78 | Helped | MAESTRO : Adaptive Sparse Attention and Robust Learning for Multimodal Dynamic Time Series (NeurIPS 2025 · 0 citations) |
| Add retrieved text to the TS-only prompt | Weather forecasting, H=24 · MAE | 0.556 -> 0.555 | No clear effect | TRACE: Grounding Time Series in Context for Multimodal Embedding and Retrieval (NeurIPS 2025 · 0 citations) |
| Add spatial semantic anchors | Synth-M · FID | 29.98±0.68 -> 29.05±0.62 | Helped | VerbalTS: Generating Time Series from Texts (ICML 2025) |
| add CT-RoPE | MIRA ablation evaluation · RMSE | 0.158 -> 0.154 | Helped | MIRA: Medical Time Series Foundation Model for Real-World Health Data (NeurIPS 2025) |
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.
Datasets
- PTB-XL · 3 of 19 papers
- ETTm1 · 2 of 19 papers
- Weather · 2 of 19 papers
- Traffic · 2 of 19 papers
- MIMIC-III · 2 of 19 papers
- MIMIC-IV · 2 of 19 papers
- TimeMMD · 2 of 19 papers
- GDELT · 2 of 19 papers
Baselines
- DLinear · 4 of 19 papers
- iTransformer · 4 of 19 papers
- Moirai · 4 of 19 papers
- TimesNet · 4 of 19 papers
- TimeLLM · 3 of 19 papers
- TimesFM · 3 of 19 papers
- Chronos · 3 of 19 papers
- PatchTST · 3 of 19 papers
Recent preprints
Loading recent arXiv preprints…
Who publishes on it
Institutions on the most papers of this approach since 2023 (of 44 papers; from OpenAlex and the papers' affiliations).
- National University of Singapore · 3 papers
- Peng Cheng Laboratory · 2 papers
- University of California, San Diego · 2 papers
- Amazon · 2 papers
- Peking University · 2 papers
- Yale University · 2 papers
Use it
Download BibTeX of 20 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 foundation models (71 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).