Research map / Label-efficient and robust learning / OOD and anomaly detection
On the Importance of Feature Separability in Predicting Out-Of-Distribution Error
RENCHUNZI XIE, Hongxin Wei, Lei Feng, Yuzhou Cao, Bo An
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 |
|---|---|---|---|
| Dispersion Score instead of Frechet on TinyImageNet | Dispersion Score · TinyImageNet avg | R^2: 0.847 -> 0.970 | Improved |
| Dispersion Score instead of ProjNorm in evaluation time (WRN-50-2) | Dispersion Score · CIFAR-10C | time (s): 575.888 -> 11.017 | Improved |
| Dispersion Score instead of ProjNorm on CIFAR-100 | Dispersion Score · CIFAR-100C avg | R^2: 0.985 -> 0.962 | Hurt |
| Dispersion Score instead of ProjNorm on CIFAR-10 | Dispersion Score · CIFAR-10C avg | R^2: 0.947 -> 0.972 | Improved |
| Dispersion Score instead of ProjNorm under class imbalance | Dispersion Score · imbalanced CIFAR-10C | R^2: 0.873 -> 0.953 | Improved |
| Dispersion Score instead of ProjNorm under class imbalance (time) | Dispersion Score · imbalanced CIFAR-10C | time (s): 398.972 -> 2.780 | Improved |
| add class-size weights to the dispersion score (ablation) | Dispersion Score · imbalanced CIFAR-10C | R^2: 0.712 -> 0.978 | Improved |
| add class-size weights to the dispersion score (ablation) | Dispersion Score · imbalanced CIFAR-100C | R^2: 0.494 -> 0.952 | Improved |
| use ground-truth labels instead of pseudo labels (ablation) | Dispersion Score · CIFAR-10 avg | R^2: 0.972 -> 0.970 | No clear effect |
| use ground-truth labels instead of pseudo labels (ablation) | Dispersion Score · CIFAR-100 avg | R^2: 0.962 -> 0.950 | No clear effect |
| use ground-truth labels for feature normalization under adversarial attack (ablation) | Dispersion Score · CIFAR-10 PGD, MSE | MSE: 1.359 -> 0.483 | Improved |
| use K-means clusters instead of classifier pseudo labels (ablation) | Dispersion Score · OOD error prediction | R^2 | Hurt |
Closest papers by meaning
- CASE: Exploiting Intra-class Compactness and Inter-class Separability of Feature Embeddings for Out-of-Distribution Detection (AAAI 2024 · similarity 0.88)
- Aggregation Hides Out-of-Distribution Generalization Failures from Spurious Correlations (NeurIPS 2025 · similarity 0.88)
- Towards Robust Out-of-Distribution Generalization Bounds via Sharpness (ICLR 2024 · similarity 0.88)
- Pursuing Feature Separation based on Neural Collapse for Out-of-Distribution Detection (ICLR 2025 · similarity 0.87)
- Feature Contamination: Neural Networks Learn Uncorrelated Features and Fail to Generalize (ICML 2024 · similarity 0.87)
- Learning to Shape In-distribution Feature Space for Out-of-distribution Detection (NeurIPS 2024 · similarity 0.86)
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.
