Research map / Label-efficient and robust learning / Recognition and long-tail
MixUp as Locally Linear Out-of-Manifold Regularization
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 |
|---|---|---|---|
| Use AdaMixUp Fold-2 instead of standard MixUp | MNIST, Fashion, SVHN, Cifar10, Cifar100, Cifar10-S, Cifar100-S, and ImageNet-R · Eight image-classification test datasets | error rate (%): AdaMixUp outperforms MixUp on all eight testing datasets | Helped |
| Apply standard MixUp with the default α | Baseline classifiers · MNIST, Cifar10-S, Cifar100-S, ImageNet-R top-1 and top-5 | error rate (%): 0.52 -> 0.57; 7.68 -> 7.88; 28.47 -> 29.39; 53.00 -> 54.89; 29.41 -> 31.02 | Hurt |
| Increase the mixing fold from 2 to 3 (ablation) | AdaMixUp · MNIST, Fashion, SVHN, Cifar10, Cifar100, Cifar10-S, Cifar100-S | error rate (%): Fold-2 -> Fold-3: 0.49 -> 0.42; 6.21 -> 5.88; 3.12 -> 2.98; 3.52 -> 3.25; 20.97 -> 20.68; 6.85 -> 6.14; 26.72 -> 25.72 | Helped |
| Increase the mixing fold from 2 to 4 (ablation) | AdaMixUp · MNIST, Fashion, SVHN, Cifar10, Cifar100, Cifar10-S, Cifar100-S | error rate (%): Fold-2 -> Fold-4: 0.49 -> 0.41; 6.21 -> 5.71; 3.12 -> 2.91; 3.52 -> 3.21; 20.97 -> 20.29; 6.85 -> 5.96; 26.72 -> 25.49 | Helped |
| Remove the intrusion discriminator (ablation) | AdaMixUp Fold-2 · Cifar10 | error rate (%): 3.52 -> 4.94 | Hurt |
| Remove the intrusion discriminator (ablation) | AdaMixUp Fold-2 · Cifar100 | error rate (%): 20.97 -> 24.75 | Hurt |
| Move interpolation to the layer before the final softmax (ablation) | AdaMixUp Fold-2 · Cifar10 | error rate (%): 3.52 -> 4.94 | Hurt |
| Move interpolation to the layer before the final softmax (ablation) | AdaMixUp Fold-2 · Cifar100 | error rate (%): 20.97 -> 22.21 | Hurt |
| Use more ResNet-18 filters (ablation) | AdaMixUp Fold-2 · Cifar100 | error rate (%) | Helped |
| Use more ResNet-18 filters (ablation) | ResNet-18 baseline · Cifar100 | error rate (%) | No clear effect |
| Increase the training sample size (ablation) | AdaMixUp Fold-2 and ResNet-18 · Cifar100 | error rate (%) | Helped |
Closest papers by meaning
- Nonlinear Mixup: Out-Of-Manifold Data Augmentation for Text Classification (AAAI 2020 · similarity 0.92)
- Tailoring Mixup to Data for Calibration (ICLR 2025 · similarity 0.88)
- Pushing Boundaries: Mixup's Influence on Neural Collapse (ICLR 2024 · similarity 0.87)
- Selective Mixup Helps with Distribution Shifts, But Not (Only) because of Mixup (ICML 2024 · similarity 0.86)
- Catch-Up Mix: Catch-Up Class for Struggling Filters in CNN (AAAI 2024 · similarity 0.86)
- Adversarial AutoMixup (ICLR 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.