Research map / Label-efficient and robust learning / OOD and anomaly detection
Removing Undesirable Feature Contributions Using Out-of-Distribution Data
Saehyung Lee, Changhwa Park, Hyungyu Lee, Jihun Yi, Jonghyun Lee, Sungroh Yoon
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 OOD data from 80M-TI with uniform labels to TRADES adversarial training | CIFAR10 TRADES · CIFAR10 | accuracy (AA): 52.83 -> 54.63 | Improved |
| Add OOD data from 80M-TI with uniform labels to TRADES adversarial training | CIFAR100 TRADES · CIFAR100 | accuracy (AA): 26.91 -> 27.83 | Improved |
| Add ImgNet990 OOD data with uniform labels to PGD training | ImgNet10 PGD · ImgNet10 (64 x 64) | accuracy (AA): 48.34 -> 53.83 | Improved |
| Add 80M-TI OOD data with uniform labels | CIFAR10 Standard · CIFAR10 (N=2,500) | accuracy: 65.44 -> 72.49 | Improved |
| Add Simpson OOD data with uniform labels | CIFAR100 Standard · CIFAR100 (N=2,500) | accuracy: 24.41 -> 27.04 | Improved |
| Add OOD data to pseudo-label training | CIFAR10 Pseudo-label · CIFAR10 (Full) | accuracy: 95.28 -> 95.53 | Improved |
| Add OOD data to pseudo-label training | CIFAR100 Pseudo-label · CIFAR100 (Full) | accuracy: 77.24 -> 77.36 | Improved |
| Increase OOD loss weight from alpha=1.0 to alpha=5.0 (ablation) | OAT · CIFAR10 | accuracy (PGD20): 57.45 -> 59.48 | Improved |
| Increase OOD loss weight from alpha=1.0 to alpha=5.0 (ablation) | OAT · CIFAR10 | accuracy (CW20): 52.65 -> 51.24 | Hurt |
| Use VisDA17 OOD data instead of no OOD augmentation | ImgNet10 PGD · ImgNet10 (64 x 64) | accuracy (PGD20): 49.00 -> 55.34 | Improved |
| Use Places365 OOD data instead of no OOD augmentation | CIFAR10 PGD · CIFAR10 | accuracy (Clean): 87.48 -> 82.37 | Hurt |
| Use OAT on full-data ImgNet10 (64 x 64), dog-breed class subset | Standard learning · ImgNet10 (64 x 64, Full) | accuracy: 75.77 -> 75.17 | Hurt |
Closest papers by meaning
- The Value of Out-of-Distribution Data (ICML 2023 · similarity 0.90)
- Improved OOD Generalization via Adversarial Training and Pretraing (ICML 2021 · similarity 0.90)
- Improving Out-of-Distribution Generalization by Adversarial Training with Structured Priors (NeurIPS 2022 · similarity 0.89)
- On the Adversarial Robustness of Out-of-distribution Generalization Models (NeurIPS 2023 · similarity 0.89)
- Understanding and Improving Feature Learning for Out-of-Distribution Generalization (NeurIPS 2023 · similarity 0.89)
- Splitting & Integrating: Out-of-Distribution Detection via Adversarial Gradient Attribution (ICML 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.
