Research map / Time series / Anomaly detection
Anomaly Transformer: Time Series Anomaly Detection with Association Discrepancy
Read the paperCheck your idea against 108,967 papers
AI-extracted summary. Written by AI from the paper's title only (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 the association-based criterion combining discrepancy and reconstruction by multiplication, rather than reconstruction alone (ablation) | Anomaly Transformer · Average across SMD, MSL, SMAP, SWaT, PSM | F1: 76.20→94.96 | Helped |
| Use association discrepancy alone rather than reconstruction alone as the anomaly criterion (ablation) | Anomaly Transformer · Average across SMD, MSL, SMAP, SWaT, PSM | F1: 76.20→91.55 | Helped |
| Use a learnable Gaussian prior-association scale instead of fixing σ at 1.0 (ablation) | Anomaly Transformer · Average across SMD, MSL, SMAP, SWaT, PSM | F1: 79.05→87.48 | Helped |
| Use minimax optimization instead of direct maximization (ablation) | Anomaly Transformer · Average across SMD, MSL, SMAP, SWaT, PSM | F1: 87.48→94.96 | Helped |
| Add Anomaly-Attention and association-based detection instead of using a pure Transformer with reconstruction (ablation) | Anomaly Transformer · Average across SMD, MSL, SMAP, SWaT, PSM | F1: 76.62→94.96 | Helped |
| Combine association discrepancy and reconstruction by multiplication rather than addition (ablation) | Anomaly Transformer · Average across SMD, MSL, SMAP, SWaT, PSM | F1: 82.32→94.96 | Helped |
| Average association discrepancy across multiple layers instead of using layer 1 alone (ablation) | Anomaly Transformer · SMD | F1: 89.92→92.33 | Helped |
| Use symmetrized KL rather than Jensen–Shannon divergence to define association discrepancy (ablation) | Anomaly Transformer · SMD | F1: 87.64→92.33 | Helped |
| Use Gaussian prior-association rather than power-law prior-association (ablation) | Anomaly Transformer · SMAP | F1: 71.31→96.69 | Helped |
| Use a Gaussian prior-association with learnable scale rather than fixed scale (ablation) | Anomaly Transformer · SMD | F1: 83.95→88.88 | Helped |
| Use minimax optimization rather than direct maximization (ablation) | Anomaly Transformer · SMD | contrast of adjacent association weights (abnormal/normal): 1.27→2.39 | Helped |
| Use direct maximization rather than reconstruction-only association learning (ablation) | Anomaly Transformer · SMD | contrast of adjacent association weights (abnormal/normal): 1.15→1.27 | Helped |
Closest papers by meaning
- SARAD: Spatial Association-Aware Anomaly Detection and Diagnosis for Multivariate Time Series (NeurIPS 2024 · similarity 0.90)
- Structured Temporal Causality for Interpretable Multivariate Time Series Anomaly Detection (NeurIPS 2025 · similarity 0.88)
- The Elephant in the Room: Towards A Reliable Time-Series Anomaly Detection Benchmark (NeurIPS 2024 · similarity 0.88)
- Position: Quo Vadis, Unsupervised Time Series Anomaly Detection? (ICML 2024 · similarity 0.87)
- When Model Meets New Normals: Test-Time Adaptation for Unsupervised Time-Series Anomaly Detection (AAAI 2024 · similarity 0.87)
- Retrieval Over Training: Similarity search-based Model Selection for Time Series Anomaly Detection (NeurIPS 2026 · 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.