Research map / Label-efficient and robust learning / OOD and anomaly detection
Rethinking Unsupervised Outlier Detection via Multiple Thresholding
Zhonghang Liu, Panzhong Lu, Guoyang Xie, Zhichao Lu, Wen-Yan Lin
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 Multi-T inlier/outlier thresholding to DeepSVDD | DeepSVDD · STL-10 (ResNet-50) | AUC: 0.622 -> 0.925 | Improved |
| add Multi-T to OCSVM | OCSVM · STL-10 (ResNet-50) | AUC: 0.927 -> 0.957 | Improved |
| add Multi-T to ABOD on STL-10 | ABOD · STL-10 (ResNet-50) | AUC: 0.665 -> 0.883 | Improved |
| add Multi-T to NeuTraL on STL-10 | NeuTraL · STL-10 (ResNet-50) | AUC: 0.854 -> 0.950 | Improved |
| add predicted inliers for manifold learning (Multi-T inliers) (ablation) | DeepSVDD + Multi-T · STL-10 (ResNet-50) | AUC: 0.643 -> 0.950 | Improved |
| add predicted outliers for Shell normalization (ablation) | DeepSVDD + Multi-T · STL-10 (ResNet-50) | AUC: 0.939 -> 0.950 | Improved |
| add predicted inliers to LVAD-S (ablation) | LVAD-S + Multi-T · STL-10 (ResNet-50) | AUC: 0.965 -> 0.971 | Improved |
| add outlier normalization to LVAD-S (ablation) | LVAD-S + Multi-T · STL-10 (ResNet-50) | AUC: 0.951 -> 0.971 | Improved |
| use Multi-T instead of classical k-sigma rule (ablation) | thresholding · STL-10 | F0.1: 0.406 -> 0.840 | Improved |
| use Multi-T instead of 1-sigma for inlier-recall metric (ablation) | thresholding · STL-10 | F10: 0.806 -> 0.869 | Improved |
| use CLIP instead of ResNet-50 features with Multi-T | Multi-T · MIT-Places | AUC: 0.867 -> 0.974 | Improved |
| add Multi-T to compress fitting data for OCSVM | OCSVM · 10,000 samples runtime | time (s): 154.8s -> 105.4s | Improved |
Closest papers by meaning
- RODEO: Robust Outlier Detection via Exposing Adaptive Out-of-Distribution Samples (ICML 2024 · similarity 0.86)
- FlexUOD: The Answer to Real-world Unsupervised Image Outlier Detection (CVPR 2025 · similarity 0.86)
- ODIM: Outlier Detection via Likelihood of Under-Fitted Generative Models (ICML 2024 · similarity 0.85)
- DOS: Diverse Outlier Sampling for Out-of-Distribution Detection (ICLR 2024 · similarity 0.85)
- Quantile-Based Maximum Likelihood Training for Outlier Detection (AAAI 2024 · similarity 0.85)
- Tight Rates in Supervised Outlier Transfer Learning (ICLR 2024 · similarity 0.85)
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.
