Research map / Label-efficient and robust learning / OOD and anomaly detection
AdaSCALE: Adaptive Scaling for OOD Detection
Sudarshan Regmi
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 Q' (Q plus correction term) instead of Q alone (ablation) | AdaSCALE-A · ImageNet-1k near-OOD avg 8 archs | FPR@95: 68.90 -> 61.17 | Improved |
| add Q' instead of Q alone (ablation) | AdaSCALE-A · ImageNet-1k far-OOD avg 8 archs | FPR@95: 42.63 -> 30.11 | Improved |
| add adaptive scaling on top of Q as OOD score (ablation) | AdaSCALE-A · ImageNet-1k near-OOD ResNet-50 | FPR@95: 79.81 -> 59.43 | Improved |
| add the correction term to lambda*Q (balanced Q') (ablation) | AdaSCALE-A · ImageNet-1k near-OOD ResNet-50 | FPR@95: 59.43 -> 58.97 | Improved |
| add Q to the correction-only OODness score (ablation) | AdaSCALE-A · ImageNet-1k near-OOD ResNet-50 | FPR@95: 70.39 -> 58.97 | Improved |
| set perturbation magnitude to 0.5 instead of 0.1 (ablation) | AdaSCALE-A · ImageNet-1k near-OOD ResNet-50 | FPR@95: 63.76 -> 58.97 | Improved |
| set perturbation magnitude to 0.5 instead of 1.0 (ablation) | AdaSCALE-A · ImageNet-1k near-OOD ResNet-50 | FPR@95: 61.60 -> 58.97 | Improved |
| set k1 to 1% instead of 100% (ablation) | AdaSCALE-A · ImageNet-1k near-OOD ResNet-50 | FPR@95: 60.99 -> 58.97 | Improved |
| set perturbed pixel fraction o to 5% instead of 100% (ablation) | AdaSCALE-A · ImageNet-1k near-OOD ResNet-50 | FPR@95: 67.37 -> 58.97 | Improved |
| use 5000 instead of 10 ID validation samples for eCDF (ablation) | AdaSCALE-A · ImageNet-1k near-OOD ResNet-50 | FPR@95: 59.69 -> 58.97 | No clear effect |
| add adaptive percentile to static scaling (AdaSCALE-A vs SCALE) | AdaSCALE-A · ImageNet-1k near-OOD ResNet-50 | FPR@95: 67.76 -> 58.98 | Improved |
| tune only p_max with p_min fixed at 60 (ablation) | AdaSCALE-A · ImageNet-1k near-OOD avg 8 archs | FPR@95: 71.91 -> 62.29 | Improved |
Earlier methods this paper compares with
Named baselines from earlier accepted papers in this paper's results tables: better than 1 of 1 on most shared results.
- ReAct — ReAct: Out-of-distribution Detection With Rectified Activations (NeurIPS 2021) · better on 1 of 1 shared result
Matched by method name in results tables read by AI. Authors' own tables tend to favour the newer method, and a comparison can be missed when a method is renamed.
Closest papers by meaning
- Scaling for Training Time and Post-hoc Out-of-distribution Detection Enhancement (ICLR 2024 · similarity 0.90)
- Leveraging Perturbation Robustness to Enhance Out-of-Distribution Detection (CVPR 2025 · similarity 0.88)
- Detecting Out-of-Distribution Through the Lens of Neural Collapse (CVPR 2025 · similarity 0.87)
- OODD: Test-time Out-of-Distribution Detection with Dynamic Dictionary (CVPR 2025 · similarity 0.87)
- Splitting & Integrating: Out-of-Distribution Detection via Adversarial Gradient Attribution (ICML 2025 · similarity 0.87)
- Enhancing the Power of OOD Detection via Sample-Aware Model Selection (CVPR 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.
