Research map / Label-efficient and robust learning / OOD and anomaly detection
Learning Non-Linear Invariants for Unsupervised Out-of-Distribution Detection
Lars Doorenbos, Raphael Sznitman, Pablo Márquez Neila
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 |
|---|---|---|---|
| replace linear (PCA) invariants with non-linear volume-preserving network invariants (ablation) | MahaAD · General U-OOD benchmark | mean AUC: 86.5 -> 87.2 | Improved |
| add backward (reconstruction) loss (ablation) | NL-Invs · General U-OOD benchmark | mean AUC: 86.9 -> 87.2 | Improved |
| add 2-NN score to invariant score (ablation) | NL-Invs · General U-OOD benchmark | mean AUC: 87.2 -> 87.9 | Improved |
| use non-linear instead of affine invariants on tabular data | NL-Invs vs MahaAD · Shallow tabular U-OOD | mean AUC: 82.9 -> 93.5 (10.6 points) | Improved |
| use non-linear invariants on deep image features vs affine | NL-Invs vs MahaAD · General U-OOD benchmark | mean AUC: 86.5 -> 87.9 | Improved |
| vary hyperparameter p from 0.5 to 10 (ablation) | NL-Invs · General U-OOD (ResNet18) | AUC: 86.5 to 86.8 | No clear effect |
| use stronger pre-trained backbone (ConvNeXT-B vs ResNet18) (ablation) | NL-Invs · uni-class | AUC: 87.8 -> 94.4 | Improved |
| NL-Invs vs best baseline on uni-ano and uni-class | NL-Invs · uni-ano / uni-class | AUC: 85.8 vs 95.7 (CFlow); 93.3 vs 96.3 (MSCL) | Hurt |
| pre-trained features for shift-low-res (CIFAR10:SVHN) | NL-Invs · shift-low-res | AUC: 97.8 | Improved |
Closest papers by meaning
- Data Invariants to Understand Unsupervised Out-of-Distribution Detection (ECCV 2022 · similarity 0.91)
- Adversarially Robust Out-of-Distribution Detection Using Lyapunov-Stabilized Embeddings (ICLR 2025 · similarity 0.88)
- MahaVar: OOD Detection via Class-wise Mahalanobis Distance Variance under Neural Collapse (NeurIPS 2026 · similarity 0.87)
- Splitting & Integrating: Out-of-Distribution Detection via Adversarial Gradient Attribution (ICML 2025 · similarity 0.87)
- Mysteries of the Deep: Role of Intermediate Representations in Out of Distribution Detection (NeurIPS 2025 · similarity 0.87)
- Leveraging Perturbation Robustness to Enhance Out-of-Distribution Detection (CVPR 2025 · 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.
