Research map / Label-efficient and robust learning / OOD and anomaly detection
Beyond Mahalanobis Distance for Textual OOD Detection
Pierre Colombo, Eduardo Dadalto Câmara Gomes, Guillaume Staerman, Nathan Noiry, Pablo Piantanida
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 |
|---|---|---|---|
| replace Mahalanobis score by IRW depth, mean aggregation (ablation) | TRUSTED · 1440-config average | AUROC: 81.2 -> 97.0 | Improved |
| replace Mahalanobis score by IRW depth, mean aggregation (ablation) | TRUSTED · 1440-config average | FPR: 40.2 -> 15.4 | Improved |
| use mean over all layers instead of last layer for IRW depth (ablation) | IRW depth · 1440-config average | AUROC: 92.6 -> 97.0 | Improved |
| use mean over all layers instead of concatenation for IRW depth (ablation) | IRW depth · 1440-config average | AUROC: 96.1 -> 97.0 | Improved |
| use mean over all layers instead of last layer for Mahalanobis (ablation) | Mahalanobis · 1440-config average | AUROC: 93.8 -> 81.2 | Hurt |
| concatenate all layers for Mahalanobis instead of last layer (ablation) | Mahalanobis · 1440-config average | AUROC: 93.8 -> 90.4 | Hurt |
| use training-set based scores (DM) instead of softmax scores | Energy/MSP · 1440-config average | AUROC: 89.7 -> 93.8 | Improved |
| use RoBERTa encoder instead of BERT | TRUSTED · per-encoder average | AUROC: 97.4 -> 96.3 | Hurt |
| use RoBERTa encoder | Mahalanobis · per-encoder average | AUROC: 95.9 -> 91.7 | Hurt |
| replace Mahalanobis by TRUSTED on SST2 in-distribution | TRUSTED · SST2 | FPR: 43.0 -> 30.7 | Improved |
| replace Mahalanobis by TRUSTED on TREC in-distribution | TRUSTED · TREC | AUROC: 99.0 -> 97.6 | Hurt |
| replace Mahalanobis by TRUSTED on IMDB | TRUSTED · IMDB | AUROC: 93.3 -> 98.6 | Improved |
Closest papers by meaning
- Unsupervised Layer-Wise Score Aggregation for Textual OOD Detection (AAAI 2024 · similarity 0.89)
- The Best of Both Worlds: On the Dilemma of Out-of-distribution Detection (NeurIPS 2024 · similarity 0.89)
- What If the Input is Expanded in OOD Detection? (NeurIPS 2024 · similarity 0.88)
- Fast Decision Boundary based Out-of-Distribution Detector (ICML 2024 · similarity 0.87)
- Test-Time Linear Out-of-Distribution Detection (CVPR 2024 · similarity 0.87)
- Beyond Clean Training Data: A Versatile and Model-Agnostic Framework for Out-of-Distribution Detection with Contaminated Training Data (CVPR 2025 · similarity 0.86)
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.
