Research map / Time series / Anomaly detection
Towards Foundation Models for Zero-Shot Time Series Anomaly Detection: Leveraging Synthetic Data and Relative Context Discrepancy
Tian Lan, Hao Duong Le, Jinbo Li, Wenjun He, Meng Wang, Chenghao Liu, Chen Zhang
Read the paperCheck your idea against 137,217 papers
AI-extracted. The measured effects below were read by AI from the paper's full text (no summary yet). Numbers and claims can be wrong or missing context, so check the paper before you cite or build on it.
Measured effects reported in the paper
| What was changed | Where | Measured | Effect |
|---|---|---|---|
| use anomaly head instead of reconstruction-error inference (ablation) | TimeRCD · UCR | VUS-PR: 4.08 -> 33.57 | Improved |
| add anomaly head supervision to reconstruction-only training (ablation) | TimeRCD · UCR | VUS-PR: 3.35 -> 33.57 | Improved |
| add masked-reconstruction auxiliary loss to anomaly-only training (ablation) | TimeRCD · UCR | VUS-PR: 5.90 -> 33.57 | Improved |
| jointly train encoder instead of frozen reconstruction-pretrained encoder (ablation) | TimeRCD · UCR | VUS-PR: 26.11 -> 33.57 | Improved |
| use synthetic data with in-context anomaly injection instead of real data with DADA injection (ablation) | TimeRCD · univariate benchmarks | F1-T: 0.073 -> 0.569 | Improved |
| use in-context injection instead of DADA injection on synthetic series (ablation) | TimeRCD · univariate benchmarks | F1-T: 0.505 -> 0.569 | Improved |
| use in-context injection instead of DADA injection on synthetic series (ablation) | TimeRCD · univariate benchmarks | VUS-PR: 0.478 vs 0.487 | No clear effect |
| scale synthetic pretraining data from 350M to 2.5B points (ablation) | TimeRCD · avg VUS-PR across benchmarks | VUS-PR | Improved |
| increase context window from 1k to 13k (ablation) | TimeRCD · UCR, Power, SMAP, SMD | detection performance | Improved |
| increase context window (ablation) | TimeRCD · YAHOO | detection performance | No clear effect |
| use TimeRCD instead of foundation baselines (zero-shot) | TimeRCD · 16 datasets x 4 metrics | rank: first in 38 and second in 14 of 64 cases | Improved |
| use large context for contextual anomalies | TimeRCD · synthetic contextual-anomaly split | Standard-F1: 0.827 vs competitors degrade | Improved |
Earlier methods this paper compares with
Named baselines from earlier accepted papers in this paper's results tables: better than 1 of 1 on most shared results.
- MOMENT — MOMENT: A Family of Open Time-series Foundation Models (ICML 2024) · better on 3 of 3 shared results
Matched by method name in results tables read by AI. Authors' own tables tend to favour the newer method, and a comparison can be missed when a method is renamed.
Closest papers by meaning
- Supervision Recovery for Time Series Anomaly Detection via Counterfactual Pairing (NeurIPS 2026 · similarity 0.89)
- Detect, Explain, Interpret: An End-to-End Benchmark for Time Series Anomaly Detection, Explainability and Interpretability. (NeurIPS 2026 · similarity 0.89)
- Structured Temporal Causality for Interpretable Multivariate Time Series Anomaly Detection (NeurIPS 2025 · similarity 0.89)
- The Elephant in the Room: Towards A Reliable Time-Series Anomaly Detection Benchmark (NeurIPS 2024 · similarity 0.89)
- CrossAD: Time Series Anomaly Detection with Cross-scale Associations and Cross-window Modeling (NeurIPS 2025 · similarity 0.88)
- Position: Quo Vadis, Unsupervised Time Series Anomaly Detection? (ICML 2024 · similarity 0.87)
Writing a paper in this area? scime draft is a free LaTeX editor with AI included, real-time co-authors and a check of your references and venue rules before you submit. It also imports OpenAI Prism projects.
