Research map / Label-efficient and robust learning / OOD and anomaly detection
AnoShift: A Distribution Shift Benchmark for Unsupervised Anomaly Detection
Marius Dragoi, Elena Burceanu, Emanuela Haller, Andrei Manolache, Florin Brad
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 |
|---|---|---|---|
| Evaluate on chronologically farther data rather than IID test data | Unsupervised anomaly detectors on Kyoto-2006+ · AnoShift IID to NEAR to FAR | ROC-AUC: 92.67 -> 87.00 -> 34.53 (deepSVDD) | Hurt |
| Use teacher-guided knowledge distillation on successive yearly data instead of IID training (ablation) | BERT for anomalies · Kyoto-2006+ 2014 | ROC-AUC: 28.56 -> 43.65 | Improved |
| Use teacher-guided knowledge distillation on successive yearly data instead of finetuning (ablation) | BERT for anomalies · Kyoto-2006+ 2015 | ROC-AUC: 53.39 -> 59.93 | Improved |
| Use teacher-guided knowledge distillation on successive yearly data instead of IID training (ablation) | BERT for anomalies · Kyoto-2006+ 2015 | ROC-AUC: 49.01 -> 59.93 | Improved |
| Use teacher-guided distillation rather than IID training (ablation) | BERT for anomalies · Kyoto-2006+ 2011 and 2012 | ROC-AUC: 88.95 -> 83.74; 95.85 -> 95.10 | Mixed |
| Do not binarize numerical features (ablation) | IsoForest · Kyoto-2006+ FAR | ROC-AUC: 27.16 -> 32.81 | Improved |
| Do not binarize numerical features (ablation) | ECOD · Kyoto-2006+ FAR | ROC-AUC: 49.19 -> 60.84 | Improved |
| Do not binarize numerical features (ablation) | COPOD · Kyoto-2006+ FAR | ROC-AUC: 50.42 -> 60.12 | Improved |
| Do not binarize numerical features (ablation) | AE for anomalies · Kyoto-2006+ FAR | ROC-AUC: 19.96 -> 30.79 | Improved |
| Do not binarize numerical features (ablation) | OC-SVM · Kyoto-2006+ IID / NEAR / FAR | ROC-AUC: 76.86 -> 81.87; 71.43 -> 71.24; 49.57 -> 50.95 | Mixed |
| Do not binarize numerical features (ablation) | IsoForest · Kyoto-2006+ IID / NEAR / FAR | ROC-AUC: 86.09 -> 94.41; 75.26 -> 95.13; 27.16 -> 32.81 | Improved |
| Do not binarize numerical features (ablation) | ECOD · Kyoto-2006+ IID / NEAR / FAR | ROC-AUC: 84.76 -> 79.38; 44.87 -> 69.80; 49.19 -> 60.84 | Mixed |
Closest papers by meaning
- Expecting The Unexpected: Towards Broad Out-Of-Distribution Detection (NeurIPS 2024 · similarity 0.86)
- Unifying Unsupervised Graph-Level Anomaly Detection and Out-of-Distribution Detection: A Benchmark (ICLR 2025 · similarity 0.86)
- We Need to Rethink Benchmarking in Anomaly Detection (NeurIPS 2026 · similarity 0.86)
- SEAD: Unsupervised Ensemble of Streaming Anomaly Detectors (ICML 2025 · similarity 0.85)
- On Diffusion Modeling for Anomaly Detection (ICLR 2024 · similarity 0.83)
- The Elephant in the Room: Towards A Reliable Time-Series Anomaly Detection Benchmark (NeurIPS 2024 · similarity 0.83)
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.
