Foundation models for time series: research map
431 papers on Foundation models for time series in Time series, from ICML, NeurIPS, ICLR, CVPR, AAAI, ICCV and ECCV, and the journals TPAMI, IJCV and JMLR (2016–2026), grouped into 2 clusters and 6 approaches. The busiest year so far is 2026.
What it is
Large models pretrained on many time series, or adapted from language models, that forecast or classify new series with little or no task-specific training. The topic also covers time-series representation learning and classification.
Terms you will see on this page:
- Foundation model
- a large model pretrained broadly and reused for many tasks.
- Zero-shot forecasting
- forecasting a new series without training on it.
- Time series classification
- assigning a label to a whole series.
Start reading with the most cited papers here:
- TS2Vec: Towards Universal Representation of Time Series (AAAI 2022 · 783 citations)
- UniRepLKNet: A Universal Perception Large-Kernel ConvNet for Audio Video Point Cloud Time-Series and Image Recognition (CVPR 2024 · 328 citations)
Within Time series, its share grew from 28.2% in 2023–24 to 32.3% in 2025–26 (107 → 254 papers at ICML, NeurIPS, ICLR, CVPR and AAAI, the venues with data for all four years).
Explore Foundation models for time series in the interactive map
Working on something in this topic? Describe your idea in scime atlas to see which approach it falls under, the closest papers by meaning and how crowded the spot has become.
What changed recently
Approaches whose share of the topic moved most between 2023–24 and 2025–26 (ICML, NeurIPS, ICLR, CVPR and AAAI):
- Time series foundation models rising: 6% → 25% of the topic (6 → 64 papers)
- Time-series language reasoning rising: 5% → 15% of the topic (5 → 39 papers)
- LLM-based time series forecasting rising: 19% → 28% of the topic (20 → 70 papers)
- Time series representation learning falling: 40% → 19% of the topic (43 → 49 papers)
- Satellite image time series falling: 14% → 3% of the topic (15 → 7 papers)
Papers from 2025–2026 that are already cited most:
- TimeCMA: Towards LLM-Empowered Multivariate Time Series Forecasting via Cross-Modality Alignment (AAAI 2025 · 96 citations)
- CALF: Aligning LLMs for Time Series Forecasting via Cross-modal Fine-Tuning (AAAI 2025 · 66 citations)
- ChatTime: A Unified Multimodal Time Series Foundation Model Bridging Numerical and Textual Data (AAAI 2025 · 28 citations)
- Federated Foundation Models on Heterogeneous Time Series (AAAI 2025 · 12 citations)
Approaches and key papers
Time series representation and classification · 223 papers
Keywords: contrastive, classification, supervised
Approaches in this cluster:
- Time series representation learning (123 papers)
Learn time series embeddings with contrastive and non-contrastive objectives and augmentation. - Time series classification (70 papers)
Classify series with shapelets, multiple-instance learning and temporal-frequency co-training. - Satellite image time series (30 papers)
Classify and analyze satellite image sequences with datasets and temporal encoders.
Most cited and most cited since 2024:
- TS2Vec: Towards Universal Representation of Time Series (AAAI 2022 · 783 citations)
- UniRepLKNet: A Universal Perception Large-Kernel ConvNet for Audio Video Point Cloud Time-Series and Image Recognition (CVPR 2024 · 328 citations)
- TimesURL: Self-Supervised Contrastive Learning for Universal Time Series Representation Learning (AAAI 2024 · 107 citations)
- Medformer: A Multi-Granularity Patching Transformer for Medical Time-Series Classification (NeurIPS 2024 · 53 citations)
Time series foundation models and LLMs · 208 papers
Keywords: shot, series foundation, llm
Approaches in this cluster:
- LLM-based time series forecasting (90 papers)
Repurpose language models for forecasting through prompting, retrieval and alignment. - Time series foundation models (73 papers)
Adapt pretrained time-series models for zero-shot forecasting with in-context tuning and adapters. - Time-series language reasoning (45 papers)
Combine time series and text for clinical reasoning and multimodal embedding.
Most cited and most cited since 2024:
- One Fits All: Power General Time Series Analysis by Pretrained LM (NeurIPS 2023 · 156 citations)
- GPT4MTS: Prompt-based Large Language Model for Multimodal Time-series Forecasting (AAAI 2024 · 97 citations)
- TimeCMA: Towards LLM-Empowered Multivariate Time Series Forecasting via Cross-Modality Alignment (AAAI 2025 · 96 citations)
- CALF: Aligning LLMs for Time Series Forecasting via Cross-modal Fine-Tuning (AAAI 2025 · 66 citations)
