Research map / Label-efficient and robust learning / OOD and anomaly detection
Unsupervised Out-of-Distribution Detection by Maximum Classifier Discrepancy
Qing Yu, Kiyoharu Aizawa
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 |
|---|---|---|---|
| use unlabeled-data discrepancy training instead of ODIN | two-head discrepancy OOD detector · CIFAR-100 ID, TINr OOD (DenseNet-BC) | Detection error (%): 17.2 -> 2.0 | Improved |
| use unlabeled-data discrepancy training instead of ELOC ensemble | two-head discrepancy OOD detector · CIFAR-100 ID, TINr OOD (WRN-28-10) | Detection error (%): 9.1 -> 2.3 | Improved |
| shrink unlabeled set from 9k ID + 9k OOD to 1k ID + 1k OOD (ablation) | two-head discrepancy OOD detector · CIFAR-100 ID, TINc OOD | Detection error (%): 0.7 -> 1.5 | Hurt |
| reduce unlabeled OOD samples from 2k to 500 with 9k ID (ablation) | two-head discrepancy OOD detector · CIFAR-100 ID, TINc OOD | Detection error (%): 0.3 -> 3.8 | Hurt |
| train unlabeled OOD on TINc, test on different LSUNc (ablation) | two-head discrepancy OOD detector · CIFAR-100 ID | Detection error (%): 0.2 -> 0.5 | Hurt |
| train unlabeled OOD on LSUNc, test on TINc (ablation) | two-head discrepancy OOD detector · CIFAR-100 ID | Detection error (%): 0.2 -> 0.7 | Hurt |
| use fine-tuning for discrepancy loss (limitation) | two-head discrepancy OOD detector · classification of ID data | accuracy drop: 5% drop | Hurt |
Closest papers by meaning
- Exploiting Discrepancy in Feature Statistic for Out-of-Distribution Detection (AAAI 2024 · similarity 0.90)
- Deep Hybrid Models for Out-of-Distribution Detection (CVPR 2022 · similarity 0.89)
- Splitting & Integrating: Out-of-Distribution Detection via Adversarial Gradient Attribution (ICML 2025 · similarity 0.89)
- CASE: Exploiting Intra-class Compactness and Inter-class Separability of Feature Embeddings for Out-of-Distribution Detection (AAAI 2024 · similarity 0.89)
- ProHOC: Probabilistic Hierarchical Out-of-Distribution Classification via Multi-Depth Networks (CVPR 2025 · similarity 0.88)
- How Does Unlabeled Data Provably Help Out-of-Distribution Detection? (ICLR 2024 · 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.
