Research map / Label-efficient and robust learning / OOD and anomaly detection
Using Mixup as a Regularizer Can Surprisingly Improve Accuracy & Out-of-Distribution Robustness
Read the paperCheck your idea against 108,967 papers
AI-extracted summary. Written by AI from the paper's title only (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 clean cross-entropy term to Mixup (RegMixup) instead of Mixup only | WRN28-10 · CIFAR-100 test | accuracy (%): 82.60 -> 83.25 | Helped |
| add clean cross-entropy term to Mixup (RegMixup) | WRN28-10 · CIFAR-10 vs SVHN OOD detection | AUROC (%): 87.53 -> 96.72 | Helped |
| add clean cross-entropy term to Mixup (RegMixup) | WRN28-10 · CIFAR-10-C | accuracy (%): 81.68 -> 83.13 | Helped |
| use Mixup instead of standard cross-entropy | WRN28-10 · CIFAR-10 vs SVHN OOD detection | AUROC (%): 96.00 -> 87.53 | Hurt |
| use Mixup instead of standard cross-entropy | WRN28-10 · CIFAR-10-C | accuracy (%): 76.60 -> 81.68 | Helped |
| RegMixup instead of AugMix | WRN28-10 · CIFAR-10-C | accuracy (%): 90.02 -> 83.13 | Hurt |
| RegMixup instead of AugMix | WRN28-10 · CIFAR-10.2 | accuracy (%): 85.9 -> 88.05 | Helped |
| RegMixup instead of Deep Ensembles (5x) | WRN28-10 · CIFAR-100 test | accuracy (%): 83.85 -> 83.25 | Hurt |
| RegMixup instead of Mixup | ResNet50 · ImageNet-1K val | accuracy (%): 77.15 -> 77.68 | Helped |
| RegMixup instead of Mixup | ResNet50 · ImageNet-A | accuracy (%): 3.29 -> 5.96 | Helped |
| RegMixup instead of Deep Ensembles | ResNet50 · ImageNet-O OOD detection | AUROC (%): 53.29 -> 57.05 | Helped |
| use large alpha (10 or 20) with clean CE term (ablation) | RegMixup · CIFAR cross-validation | accuracy | Helped |
Closest papers by meaning
- Noisy Feature Mixup (ICLR 2022 · similarity 0.89)
- On the Pitfall of Mixup for Uncertainty Calibration (CVPR 2023 · similarity 0.88)
- Catch-Up Mix: Catch-Up Class for Struggling Filters in CNN (AAAI 2024 · similarity 0.86)
- Selective Mixup Helps with Distribution Shifts, But Not (Only) because of Mixup (ICML 2024 · similarity 0.86)
- Adversarial AutoMixup (ICLR 2024 · similarity 0.85)
- Cross-regularization: Adaptive Model Complexity through Validation Gradients (ICML 2025 · similarity 0.85)
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.