Research map / Label-efficient and robust learning / OOD and anomaly detection
Unsupervised Anomaly Detection with Rejection
Lorenzo Perini, Jesse Davis
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 |
|---|---|---|---|
| Set rejection using ExCeeD stability confidence and a constant threshold instead of the tested baselines | Unsupervised anomaly detectors · 34 benchmark datasets, averaged over detectors | cost per example: 0.121 ± 0.133 vs 0.125 ± 0.135 (SS-Repen); 0.148 ± 0.144 (NoReject) | Improved |
| Use ExCeeD-based rejection rather than metric-based threshold selection | Unsupervised anomaly detectors · 34 benchmark datasets | cost per example: more than 5% vs SS-Repen; 11% vs Ens; 13% vs Mv and Udr; 17% vs Em; 19% vs NoReject | Improved |
| Use RejEx rather than the baselines when false positives are costly | Unsupervised anomaly detectors · 34 benchmark datasets; false positive cost = 10 | cost per example: 11 out of 12 detectors | Improved |
| Use RejEx rather than the baselines when false negatives are costly | Unsupervised anomaly detectors · 34 benchmark datasets; false negative cost = 10 | cost per example: 11 detectors | Improved |
| Use RejEx rather than the baselines when rejection cost is low | Unsupervised anomaly detectors · 34 benchmark datasets; false positive cost = 5, false negative cost = 5, rejection cost = γ | cost per example: all detectors | Improved |
| Replace label-free thresholding with label-optimized thresholding | ExCeeD confidence · 34 benchmark datasets, averaged over detectors | cost per example: 0.121 ± 0.133 vs 0.121 ± 0.133; 0.6% higher cost | No clear effect |
| Use RejEx instead of NoReject | IForest · 34 datasets | CPU time per training example (ms): 0.0±0.0 -> 0.06±0.22 | Hurt |
| Use RejEx instead of threshold-search baselines | IForest · 34 datasets | CPU time per training example (ms): 90±68 (SS-Repen); 916±900 (Stability); RejEx: 0.06±0.22 | Improved |
Closest papers by meaning
- Interpreting Unsupervised Anomaly Detection in Security via Rule Extraction (NeurIPS 2023 · similarity 0.86)
- Online Adaptive Anomaly Thresholding with Confidence Sequences (ICML 2024 · similarity 0.85)
- We Need to Rethink Benchmarking in Anomaly Detection (NeurIPS 2026 · similarity 0.85)
- SEAD: Unsupervised Ensemble of Streaming Anomaly Detectors (ICML 2025 · similarity 0.85)
- Dissect Black Box: Interpreting for Rule-Based Explanations in Unsupervised Anomaly Detection (NeurIPS 2024 · similarity 0.84)
- Towards a Unified Framework of Clustering-based Anomaly Detection (ICML 2025 · similarity 0.84)
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.
