Research map / Time series / Foundation models for time series
Time series classification
Classify series with shapelets, multiple-instance learning and temporal-frequency co-training.
61 accepted papers at ICML, NeurIPS, ICLR, CVPR and AAAI (2016–2026) · an approach in Foundation models for time series, keyword cluster “contrastive · classification · supervised”
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 25 in 2025–26 (at ICML, NeurIPS, ICLR, CVPR and AAAI); its share of the topic shrank, from 16.8% to 9.8%.
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 · Graphical Time Warping for Joint Alignment of Multiple Curves (NeurIPS 2016 · 6 citations)
- 2018 · Learning a Warping Distance from Unlabeled Time Series Using Sequence Autoencoders (NeurIPS 2018 · 24 citations)
- 2019 · Learning Representations for Time Series Clustering (NeurIPS 2019 · 98 citations)
- 2020 · TapNet: Multivariate Time Series Classification with Attentional Prototypical Network (AAAI 2020 · 339 citations)
- 2021 · ShapeNet: A Shapelet-Neural Network Approach for Multivariate Time Series Classification (AAAI 2021 · 162 citations)
- 2022 · BILCO: An Efficient Algorithm for Joint Alignment of Time Series (NeurIPS 2022 · 0 citations)
- 2023 · Temporal-Frequency Co-training for Time Series Semi-supervised Learning (AAAI 2023 · 37 citations)
- 2024 · Medformer: A Multi-Granularity Patching Transformer for Medical Time-Series Classification (NeurIPS 2024 · 53 citations)
- 2025 · Hierarchical Classification Auxiliary Network for Time Series Forecasting (AAAI 2025 · 7 citations)
- 2026 · EdgeMTSC: A Lightweight Large-Kernel ConvNet for Multivariate Time Series Classification (AAAI 2026 · 1 citation)
Earliest papers here
- Graphical Time Warping for Joint Alignment of Multiple Curves (NeurIPS 2016 · 6 citations)
- Learning a Warping Distance from Unlabeled Time Series Using Sequence Autoencoders (NeurIPS 2018 · 24 citations)
- Learning filter widths of spectral decompositions with wavelets (NeurIPS 2018 · 16 citations)
What has been tried
46 papers here report 469 measured changes (a component added, removed or swapped, compared with a baseline): helped 77%, no clear effect 5%, hurt 12%, mixed 6%. Papers mostly publish changes that helped, so "helped" is over-represented.
Ablations in recent papers
| What was changed | Measured on | Result | Effect | Paper |
|---|---|---|---|---|
| add inter-granularity self-attention | Five datasets · F1 score | 3.64% | Helped | Medformer: A Multi-Granularity Patching Transformer for Medical Time-Series Classification (NeurIPS 2024 · 53 citations) |
| add self-supervised diffusion learning | 12 UCR datasets, 10% labeled · average rank | 3.08 -> 1.00 | Helped | Diffusion Language-Shapelets for Semi-supervised Time-Series Classification (AAAI 2024 · 31 citations) |
| replace the reservoir kernel with a Gaussian kernel | Homburg · accuracy | 82.45 -> 93.00 | Helped | Audio Scanning Network: Bridging Time and Frequency Domains for Audio Classification (AAAI 2024 · 10 citations) |
| add fine-grained UAC loss | Weather and Solar Wind · MSE | Weather: 0.352 -> 0.349, 0.636 -> 0.509, 0.680 -> 0.613, 1.265 -> 0.993; Solar Wind: 1.710 -> 0.991, 1.991 -> 1.077, 1.9 | Helped | Hierarchical Classification Auxiliary Network for Time Series Forecasting (AAAI 2025 · 7 citations) |
| Add self-supervised pre-training | Inpatient · F1-score | 0.73 -> 0.79 | Helped | Pharmacokinetics-Informed Neural Network for Predicting Opioid Administration Moments with Wearable Sensors (AAAI 2024 · 7 citations) |
| Add agreement loss LA | SleepEDF 2sub · ACC | 69.68 -> 73.21 | Helped | Self-Supervised Framework Based on Subject-Wise Clustering for Human Subject Time Series Data (AAAI 2024 · 7 citations) |
| Remove the Temporal Convolutional Network | HAR-6 · accuracy | Hurt | Energy Efficient Streaming Time Series Classification with Attentive Power Iteration (AAAI 2024 · 5 citations) | |
| combine sequence and image representations | PAM · F1 score | increasing by 2.6% and 4.5% | Helped | Integrating Sequence and Image Modeling in Irregular Medical Time Series Through Self-Supervised Learning (AAAI 2025 · 3 citations) |
| Replace mean pooling with time-aware MIL pooling | Group 1 UEA datasets (26 datasets) · Accuracy | 0.715 -> 0.774 | Helped | TimeMIL: Advancing Multivariate Time Series Classification via a Time-aware Multiple Instance Learning (ICML 2024 · 3 citations) |
| Replace CPG-PE with a random spike matrix | Text classification (6-benchmark average) · accuracy | 79.41 -> 77.23 | Hurt | Advancing Spiking Neural Networks for Sequential Modeling with Central Pattern Generators (NeurIPS 2024 · 3 citations) |
| Add language-guided masking rather than self-attention-based adaptive masking | 12 UCR time series datasets · average accuracy | 0.851 -> 0.882 | Helped | Language Pre-training Guided Masking Representation Learning for Time Series Classification (AAAI 2025 · 2 citations) |
| include positional encoding | UCR datasets · predictive performance | Mixed | Inherently Interpretable Time Series Classification via Multiple Instance Learning (ICLR 2024 · 2 citations) |
What reviewers will expect
Datasets and baselines named in the 16 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 · 2 of 16 papers
Baselines
- TS-TCC · 4 of 16 papers
- PatchTST · 3 of 16 papers
- T-Loss · 2 of 16 papers
- TST · 2 of 16 papers
- TS2Vec · 2 of 16 papers
- TimesNet · 2 of 16 papers
- TSLANet · 2 of 16 papers
- Shapeformer · 2 of 16 papers
Recent preprints
Loading recent arXiv preprints…
Who publishes on it
Institutions on the most papers of this approach since 2023 (of 43 papers; from OpenAlex and the papers' affiliations).
- South China University of Technology · 4 papers
- Microsoft · 3 papers
- Microsoft Research Asia · 3 papers
- Amazon · 2 papers
- University of California, San Diego · 2 papers
- Dalian University of Technology · 2 papers
- Agency for Science · 2 papers
- Zhejiang University · 2 papers
- UC Santa Barbara · 2 papers
- Clemson University · 2 papers
Use it
Download BibTeX of 38 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
- Time series representation learning (115 papers)
- Satellite image time series (28 papers)
Elsewhere in the topic
- LLM-based time series forecasting (90 papers)
- Time series foundation models (71 papers)
- Time-series language reasoning (44 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).