Research map / Label-efficient and robust learning / OOD and anomaly detection
Extremely Simple Activation Shaping for Out-of-Distribution Detection
Andrija Djurisic, Nebojsa Bozanic, Arjun Ashok, Rosanne Liu
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 |
|---|---|---|---|
| Prune low-valued activations and binarize the retained activations | ResNet-50 with ASH-B@65 · ImageNet OOD average | FPR95: 27.25 -> 22.73 | Improved |
| Prune low-valued activations and exponentially scale the retained activations | DenseNet-101 · CIFAR-10 OOD average | FPR95: 20.83 -> 15.05 | Improved |
| Prune low-valued activations and exponentially scale the retained activations | DenseNet-101 · CIFAR-100 OOD average | FPR95: 49.72 -> 41.40 | Improved |
| Add ASH treatment to the Energy score (ablation) | ResNet-50 · ImageNet OOD average | FPR95: 57.47 -> 22.80 | Improved |
| Add ASH treatment to ODIN (ablation) | ResNet-50 · ImageNet OOD average | FPR95: 50.80 -> 28.59 | Improved |
| Add ASH treatment to ReAct (ablation) | ResNet-50 · ImageNet OOD average | FPR95: 31.43 -> 24.88 | Improved |
| Replace exponential scaling with linear scaling (ablation) | ASH-S@90 on ResNet-50 · ImageNet OOD average | FPR95: 22.80 -> 36.72 | Hurt |
| Use a global threshold instead of a local, per-input threshold (ablation) | ASH-S@90 on ResNet-50 · ImageNet OOD average | FPR95: 22.80 -> 54.74 | Hurt |
| Move ASH-S from the penultimate layer to before the last ReLU of the 3rd Block (ablation) | ResNet-50 · ImageNet OOD average | FPR95: 22.80 -> 98.51 | Hurt |
| Add scaling to pruning-only ASH-P (ablation) | ASH-S on ResNet-50 · ImageNet OOD average | FPR95: 50.32 -> 22.80 | Improved |
| Randomize retained activations instead of binarizing or scaling them (ablation) | ASH-RAND@65 on ResNet-50 · ImageNet OOD average | FPR95: 22.73 -> 45.37 | Hurt |
| Increase pruning from 65% to 90% with ASH-S (ablation) | ResNet-50 · ImageNet validation | Top-1 accuracy: 76.01\% -> 74.98\% | Hurt |
Closest papers by meaning
- Scaling for Training Time and Post-hoc Out-of-distribution Detection Enhancement (ICLR 2024 · similarity 0.92)
- Splitting & Integrating: Out-of-Distribution Detection via Adversarial Gradient Attribution (ICML 2025 · similarity 0.91)
- Towards Optimal Feature-Shaping Methods for Out-of-Distribution Detection (ICLR 2024 · similarity 0.90)
- Enhancing the Power of OOD Detection via Sample-Aware Model Selection (CVPR 2024 · similarity 0.90)
- Mysteries of the Deep: Role of Intermediate Representations in Out of Distribution Detection (NeurIPS 2025 · similarity 0.90)
- Leveraging Perturbation Robustness to Enhance Out-of-Distribution Detection (CVPR 2025 · similarity 0.89)
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.
