Research map / Label-efficient and robust learning / OOD and anomaly detection
Activation Subspaces for Out-of-Distribution Detection
Barış Zöngür, Robin Hesse, Stefan Roth
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 insignificant-subspace cosine score to decisive shaped-energy score | SCALE-based decisive score · ImageNet-1k ResNet-50, 4 OOD sets | AUC: 95.71 -> 96.74 | Improved |
| add ActSub to SCALE | SCALE · OpenOOD Near-OOD | AUC: 81.36 -> 84.24 | Improved |
| add ActSub to ASH-S | ASH-S · OpenOOD Near-OOD | AUC: 79.63 -> 84.06 | Improved |
| add ActSub to ReAct | ReAct · OpenOOD Near-OOD | AUC: 77.38 -> 77.70 | Improved |
| add ActSub to SCALE | SCALE · CIFAR10 ID DenseNet-101 | AUC: 96.16 -> 96.87 | Improved |
| add ActSub to SCALE | SCALE · CIFAR100 ID DenseNet-101 | FPR@95TPR: 45.86 -> 41.26 | Improved |
| add ActSub to SCALE | SCALE · MobileNetV2 ImageNet-1k | FPR@95TPR: 34.18 -> 21.70 | Improved |
| use SVD basis instead of PCA basis for the combined score (ablation) | ActSub · OpenOOD Near-OOD | AUC: 80.40 -> 84.24 | Improved |
| use SVD basis instead of Null Space basis for the combined score (ablation) | ActSub · OpenOOD Near-OOD | AUC: 80.04 -> 84.24 | Improved |
| use SVD basis instead of PCA for decisive-only score (ablation) | decisive shaped score · OpenOOD Far-OOD | AUC: 96.10 -> 93.58 | Hurt |
| compute cosine score on insignificant component instead of full activation (ablation) | insignificant cosine score · OpenOOD Far-OOD | AUC: 87.77 -> 92.74 | Improved |
| compute cosine score on insignificant component instead of full activation (ablation) | insignificant cosine score · OpenOOD Near-OOD | AUC: 71.75 -> 67.51 | Hurt |
Closest papers by meaning
- Learning to Shape In-distribution Feature Space for Out-of-distribution Detection (NeurIPS 2024 · similarity 0.89)
- Scaling for Training Time and Post-hoc Out-of-distribution Detection Enhancement (ICLR 2024 · similarity 0.89)
- Detecting Out-of-Distribution Through the Lens of Neural Collapse (CVPR 2025 · similarity 0.89)
- Mysteries of the Deep: Role of Intermediate Representations in Out of Distribution Detection (NeurIPS 2025 · similarity 0.89)
- Splitting & Integrating: Out-of-Distribution Detection via Adversarial Gradient Attribution (ICML 2025 · similarity 0.89)
- Leveraging Perturbation Robustness to Enhance Out-of-Distribution Detection (CVPR 2025 · similarity 0.88)
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.
