Research map / Label-efficient and robust learning / OOD and anomaly detection
Principled Out-of-Distribution Detection via Multiple Testing
Akshayaa Magesh, Venugopal V. Veeravalli, Anirban Roy, Susmit Jha
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 |
|---|---|---|---|
| Combine Mahalanobis, Gram and Energy scores with BH-inspired test instead of using Mahalanobis alone (ablation) | Mahalanobis baseline · CIFAR10 in-dist, SVHN OOD, ResNet34 | detection power at PF=10% (%): 82.77 -> 97.13 | Improved |
| Combine all scores with BH-inspired test instead of Energy alone (ablation) | Energy baseline · CIFAR10 in-dist, SVHN OOD, DenseNet | detection power at PF=10% (%): 42.40 -> 94.57 | Improved |
| Use BH-inspired combination instead of naive averaging (ablation) | Naive averaging · SVHN in-dist, CIFAR10 OOD, DenseNet | detection power at PF=10% (%): 77.22 -> 95.23 | Improved |
| Use BH-inspired combination instead of Bonferroni combination (ablation) | Bonferroni combination · SVHN in-dist, CIFAR10 OOD, DenseNet | detection power at PF=10% (%): 91.77 -> 95.23 | Improved |
| Use BH-inspired combination instead of Bonferroni (ablation) | Bonferroni combination · CIFAR10 in-dist, SVHN OOD, ResNet34 | detection power at PF=10% (%): 96.41 -> 97.13 | Improved |
| Add Gram and Energy scores to Mahalanobis in the combination (ablation) | Ours - Mahala (5/4) · SVHN in-dist, CIFAR10 OOD, ResNet34 | detection power at PF=10% (%): 98.31 -> 97.35 | Hurt |
| Add Mahalanobis scores to Gram combination (ablation) | Ours - Gram (5/4) · SVHN in-dist, CIFAR10 OOD, DenseNet | detection power at PF=10% (%): 72.84 -> 95.23 | Improved |
| Add Energy score to Gram combination (ablation) | Ours - Gram (5/4) · SVHN in-dist, CIFAR10 OOD, DenseNet | detection power at PF=10% (%): 72.84 -> 78.03 | Improved |
| Add Energy score to Mahalanobis combination (ablation) | Ours - Mahala (5/4) · CIFAR10 in-dist, ImageNet OOD, DenseNet | detection power at PF=10% (%): 95.19 -> 98.09 | Improved |
| Add all three score types (ablation) | Mahalanobis-only combination · CIFAR10 in-dist, ImageNet OOD, DenseNet | detection power at PF=10% (%): 95.19 -> 97.20 | Improved |
| Combine all scores with BH test across datasets | Single-score baselines · CIFAR10 in-dist, ResNet34, 4 OOD sets | detection power at PF=10% (%): 82.77 - 90.97 (Mahala) to 97.03 - 98.00 | Improved |
| Combine all scores with BH test across datasets | Single-score baselines · CIFAR10 in-dist, DenseNet, 4 OOD sets | detection power at PF=10% (%): 42.40 - 96.89 (Energy) to 94.57 - 97.78 | Improved |
Closest papers by meaning
- What If the Input is Expanded in OOD Detection? (NeurIPS 2024 · similarity 0.91)
- Expecting The Unexpected: Towards Broad Out-Of-Distribution Detection (NeurIPS 2024 · similarity 0.91)
- Test-Time Linear Out-of-Distribution Detection (CVPR 2024 · similarity 0.90)
- The Best of Both Worlds: On the Dilemma of Out-of-distribution Detection (NeurIPS 2024 · similarity 0.90)
- Double Descent Meets Out-of-Distribution Detection: Theoretical Insights and Empirical Analysis on the Role of Model Complexity (NeurIPS 2025 · similarity 0.90)
- Enhancing the Power of OOD Detection via Sample-Aware Model Selection (CVPR 2024 · similarity 0.90)
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.
