Research map / Time series / Anomaly detection
Time series anomaly detection evaluation
Evaluate and improve anomaly detection on time series with metrics, foundation models and LLMs.
47 accepted papers at ICML, NeurIPS, ICLR, CVPR and AAAI (2018–2026) · an approach in Anomaly detection, keyword cluster “anomaly · detection · anomalies”
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.
8 papers in 2023–24 and 32 in 2025–26 (at ICML, NeurIPS, ICLR, CVPR and AAAI); its share of the topic held steady, from 57.1% to 54.2%.
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
- 2018 · Early Detection of Fake News on Social Media Through Propagation Path Classification with Recurrent and Convolutional Networks (AAAI 2018 · 655 citations)
- 2021 · Time Series Anomaly Detection with Multiresolution Ensemble Decoding (AAAI 2021 · 56 citations)
- 2022 · Anomaly Transformer: Time Series Anomaly Detection with Association Discrepancy (ICLR 2022 · 242 citations)
- 2023 · Unsupervised Model Selection for Time Series Anomaly Detection (ICLR 2023 · 9 citations)
- 2024 · The Elephant in the Room: Towards A Reliable Time-Series Anomaly Detection Benchmark (NeurIPS 2024 · 51 citations)
- 2025 · KAN-AD: Time Series Anomaly Detection with Kolmogorov–Arnold Networks (ICML 2025 · 3 citations)
- 2026 · Harnessing Vision-Language Models for Time Series Anomaly Detection (AAAI 2026 · 1 citation)
Earliest papers here
- Early Detection of Fake News on Social Media Through Propagation Path Classification with Recurrent and Convolutional Networks (AAAI 2018 · 655 citations)
- Precision and Recall for Time Series (NeurIPS 2018 · 38 citations)
- Time Series Anomaly Detection with Multiresolution Ensemble Decoding (AAAI 2021 · 56 citations)
What has been tried
19 papers here report 184 measured changes (a component added, removed or swapped, compared with a baseline): helped 64%, no clear effect 7%, hurt 14%, mixed 15%. Papers mostly publish changes that helped, so "helped" is over-represented.
Ablations in recent papers
| What was changed | Measured on | Result | Effect | Paper |
|---|---|---|---|---|
| Add detrending only; omit test-time adaptation | SWaT · F1 | 0.765 -> 0.762 | No clear effect | When Model Meets New Normals: Test-Time Adaptation for Unsupervised Time-Series Anomaly Detection (AAAI 2024 · 34 citations) |
| Add backcasting supervision to forecasting supervision | ETTh2, horizon 96; look-back 720; COF harmonic 6 · MSE | 0.271312 -> 0.271028 | Helped | FITS: Modeling Time Series with $10k$ Parameters (ICLR 2024 · 17 citations) |
| Vary the property-loss weights λ1, λ2, λ3, and λ4 | Time-series · formal property metrics | No clear effect | AR-Pro: Counterfactual Explanations for Anomaly Repair with Formal Properties (NeurIPS 2024 · 4 citations) | |
| remove raw time input X | UCR · Event F1 | 0.5335 -> 0.5153 | Hurt | KAN-AD: Time Series Anomaly Detection with Kolmogorov–Arnold Networks (ICML 2025 · 3 citations) |
| Add multi-scale modeling | PSM, MSL, and GECCO (average) · VUS-PR | 0.3665 -> 0.3962 | Helped | CrossAD: Time Series Anomaly Detection with Cross-scale Associations and Cross-window Modeling (NeurIPS 2025 · 3 citations) |
| add the temporal online encoder | MSL, PSM, WADI, and SWaT · Aff-F | 0.764 -> 0.867; 0.810 -> 0.794; 0.568 -> 0.587; 0.659 -> 0.704 | Mixed | ScatterAD: Temporal-Topological Scattering Mechanism for Time Series Anomaly Detection (NeurIPS 2025 · 1 citation) |
| add ReLU to association progression to retain reduction-only changes | SMD, PSM, SWaT, and HAI · VUS-ROC | 75.16 -> 79.67; 60.21 -> 61.77; 86.75 -> 87.52; 94.93 -> 96.17 | Helped | SARAD: Spatial Association-Aware Anomaly Detection and Diagnosis for Multivariate Time Series (NeurIPS 2024 · 1 citation) |
| Add AAF and SAP | MBA, Exathlon, SMD, WADI · Avg. F1 | 36.82 -> 46.84 | Helped | When Will It Fail?: Anomaly to Prompt for Forecasting Future Anomalies in Time Series (ICML 2025 · 0 citations) |
| add adaptive bottlenecks instead of using one fixed bottleneck | SMD/MSL/SMAP (zero-shot) · Avg F1 | 74.45 -> 79.49 | Helped | Towards a General Time Series Anomaly Detector with Adaptive Bottlenecks and Dual Adversarial Decoders (ICLR 2025 · 0 citations) |
| Distill GDN to a linear model | SWaT · F1 | 0.810 -> 0.808 | No clear effect | Position: Quo Vadis, Unsupervised Time Series Anomaly Detection? (ICML 2024 · 0 citations) |
What reviewers will expect
Datasets and baselines named in the 18 papers of this approach from 2025–2026 that have a full-text summary. Leaving out the common ones is a frequent review question.
Datasets
- MSL · 4 of 18 papers
- SWaT · 4 of 18 papers
- PSM · 4 of 18 papers
- SMD · 3 of 18 papers
- SMAP · 3 of 18 papers
- Exathlon · 2 of 18 papers
- WADI · 2 of 18 papers
- TODS · 2 of 18 papers
Baselines
- Anomaly Transformer · 5 of 18 papers
- DCdetector · 4 of 18 papers
- TranAD · 4 of 18 papers
- IForest · 4 of 18 papers
- OmniAnomaly · 4 of 18 papers
- FITS · 3 of 18 papers
- ModernTCN · 3 of 18 papers
- PCA · 3 of 18 papers
Recent preprints
Loading recent arXiv preprints…
Who publishes on it
Institutions on the most papers of this approach since 2023 (of 40 papers; from OpenAlex and the papers' affiliations).
- Tsinghua University · 4 papers
- Korea Advanced Institute of Science and Technology · 3 papers
- Huawei · 3 papers
- East China Normal University · 2 papers
- Aalborg University · 2 papers
- Ecole Normale Supérieure de Paris · 2 papers
- Inria · 2 papers
Use it
Download BibTeX of 29 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
- Multivariate anomaly detection (37 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).