Research map / Time series / Anomaly detection
PaAno: Patch-Based Representation Learning for Time-Series Anomaly Detection
Jinju Park, Seokho Kang
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 |
|---|---|---|---|
| add instance normalization of patches (ablation) | PaAno · TSB-AD-U (univariate) | VUS-PR (%): 45.3 -> 53.0 | Improved |
| add instance normalization of patches (ablation) | PaAno · TSB-AD-M (multivariate) | VUS-PR (%): 33.4 -> 42.6 | Improved |
| add triplet and pretext losses (ablation) | PaAno · TSB-AD-U (univariate) | VUS-PR (%): 48.0 -> 53.0 | Improved |
| add triplet and pretext losses (ablation) | PaAno · TSB-AD-M (multivariate) | VUS-PR (%): 35.6 -> 42.6 | Improved |
| use farthest negative selection in triplet loss (ablation) | PaAno · TSB-AD-U (univariate) | VUS-PR (%): 50.9 -> 53.0 | Improved |
| use farthest negative selection in triplet loss (ablation) | PaAno · TSB-AD-M (multivariate) | VUS-PR (%): 40.2 -> 42.6 | Improved |
| use triplet loss instead of InfoNCE (ablation) | PaAno · TSB-AD-U (univariate) | VUS-PR (%): 48.3 -> 53.0 | Improved |
| use triplet loss instead of InfoNCE (ablation) | PaAno · TSB-AD-M (multivariate) | VUS-PR (%): 36.2 -> 42.6 | Improved |
| add early-stage pretext loss (ablation) | PaAno · TSB-AD-U (univariate) | VUS-PR (%): 51.1 -> 53.0 | Improved |
| add early-stage pretext loss (ablation) | PaAno · TSB-AD-M (multivariate) | VUS-PR (%): 42.2 -> 42.6 | No clear effect |
| apply pretext loss only early instead of throughout training (ablation) | PaAno · TSB-AD-U (univariate) | VUS-PR (%): 47.4 -> 53.0 | Improved |
| apply pretext loss only early instead of throughout training (ablation) | PaAno · TSB-AD-M (multivariate) | VUS-PR (%): 40.8 -> 42.6 | Improved |
Later papers that compared with this method
1 later accepted paper reports PaAno as a baseline in their results tables; 1 of them report a better number on most of the results they share with it.
- VACE: Learning Geometrically Structured Representations for Time Series Anomaly Detection (NeurIPS 2026 · better on 4 of 6 shared results)
Earlier methods this paper compares with
Named baselines from earlier accepted papers in this paper's results tables: better than 3 of 3 on most shared results.
- KAN-AD — KAN-AD: Time Series Anomaly Detection with Kolmogorov–Arnold Networks (ICML 2025) · better on 2 of 2 shared results
- AnomalyTransformer — Anomaly Transformer: Time Series Anomaly Detection with Association Discrepancy (ICLR 2022) · better on 2 of 2 shared results
- CATCH — CATCH: Channel-Aware Multivariate Time Series Anomaly Detection via Frequency Patching (ICLR 2025) · better on 1 of 1 shared result
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
- The Elephant in the Room: Towards A Reliable Time-Series Anomaly Detection Benchmark (NeurIPS 2024 · similarity 0.88)
- COMET: Codebook-based Online-adaptive Multi-scale Embedding for Time-series Anomaly Detection (NeurIPS 2026 · similarity 0.87)
- VACE: Learning Geometrically Structured Representations for Time Series Anomaly Detection (NeurIPS 2026 · similarity 0.86)
- Structured Temporal Causality for Interpretable Multivariate Time Series Anomaly Detection (NeurIPS 2025 · similarity 0.86)
- Position: Quo Vadis, Unsupervised Time Series Anomaly Detection? (ICML 2024 · similarity 0.86)
- Retrieval Over Training: Similarity search-based Model Selection for Time Series Anomaly Detection (NeurIPS 2026 · similarity 0.86)
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.
