Research map / Label-efficient and robust learning / OOD and anomaly detection
Igeood: An Information Geometry Approach to Out-of-Distribution Detection
Eduardo Dadalto Camara Gomes, Florence Alberge, Pierre Duhamel, 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 |
|---|---|---|---|
| Use Fisher-Rao distances between softmax outputs and class-conditional centroids instead of Energy | Black-Box OOD detection · Six model/dataset contexts, eight OOD datasets (average) | TNR at TPR-95%: 51.1 -> 51.6 | Improved |
| Use Fisher-Rao distances between softmax outputs and class-conditional centroids instead of Energy | Black-Box OOD detection · Six model/dataset contexts, eight OOD datasets (average) | AUROC: 85.2 -> 85.7 | Improved |
| Combine logits and low-level feature scores, including OOD-feature statistics | White-Box Igeood+ · Validation on OOD data, average across models and in-distribution datasets | TNR at TPR-95%: 81.1 -> 92.9 | Improved |
| Combine logits and low-level feature scores, including OOD-feature statistics | White-Box Igeood+ · Validation on OOD data, average across models and in-distribution datasets | AUROC: 94.8 -> 98.5 | Improved |
| Combine logits and low-level feature scores, tuned using adversarial data | White-Box Igeood · Validation on adversarial data, average across models and in-distribution datasets | TNR at TPR-95%: 77.0 -> 79.5 | Improved |
| Combine logits and low-level feature scores, tuned using adversarial data | White-Box Igeood · Validation on adversarial data, average across models and in-distribution datasets | AUROC: 92.8 -> 93.4 | Improved |
| Add intermediate-feature scores to the logits score (ablation) | White-Box Igeood · OOD detection | OOD detection performance | Improved |
| Add OOD-feature statistics to the logits and intermediate-feature scores (ablation) | White-Box Igeood+ · OOD detection with validation OOD data | OOD detection performance | Improved |
| Replace the minimum class-centroid logit distance with the sum across class centroids (ablation) | Logits-level Igeood · OOD detection | OOD detection performance | Improved |
| Replace the minimum class-conditional feature distance with the sum across classes (ablation) | Feature-level Igeood · OOD detection | OOD detection performance | No clear effect |
| Use input preprocessing to perturb the test input in the direction that increases the Igeood score (ablation) | Grey-Box Igeood · Grey-Box OOD detection | TNR at TPR-95% | Mixed |
Closest papers by meaning
- Rethinking Out-of-Distribution Detection on Imbalanced Data Distribution (NeurIPS 2024 · similarity 0.89)
- A Variational Information Theoretic Approach to Out-of-Distribution Detection (ICML 2025 · similarity 0.88)
- Fast Decision Boundary based Out-of-Distribution Detector (ICML 2024 · similarity 0.88)
- Learning to Shape In-distribution Feature Space for Out-of-distribution Detection (NeurIPS 2024 · similarity 0.87)
- Mining In-distribution Attributes in Outliers for Out-of-distribution Detection (AAAI 2025 · similarity 0.87)
- MultiOOD: Scaling Out-of-Distribution Detection for Multiple Modalities (NeurIPS 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.
