Research map / Adversarial robustness and security / Certified and empirical robustness
mixup: Beyond Empirical Risk Minimization
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 |
|---|---|---|---|
| Mix inputs and labels | ResNet-101 · ImageNet-2012 val | Top-1 error: 22.1 -> 21.5 | Helped |
| Mix inputs and labels | ResNeXt-101 64×4d · ImageNet-2012 val | Top-1 error: 20.4 -> 19.8 | Helped |
| Apply mixup with large α | CIFAR-10 with corrupted labels · 20%, 50%, and 80% label corruption | Best test error: 12.7 -> 5.9; 18.8 -> 11.3; 36.5 -> 25.3 | Helped |
| Combine mixup with dropout | CIFAR-10 with corrupted labels · 20%, 50%, and 80% label corruption | Best test error: 12.7 -> 6.2; 18.8 -> 10.9; 36.5 -> 24.0 | Helped |
| Mix inputs and labels | ResNet-101 · ImageNet-2012 adversarial white-box FGSM | Top-1 error: 90.7 -> 75.2 | Helped |
| Mix inputs and labels | ResNet-101 · ImageNet-2012 adversarial black-box I-FGSM | Top-1 error: 57.3 -> 40.9 | Helped |
| Use random pairs across all classes instead of nearest neighbors (ablation) | mixup on PreAct ResNet-18 · CIFAR-10 | Median test error, last 10 epochs: 4.98 -> 4.24 | Helped |
| Add interpolated labels (ablation) | Input interpolation on PreAct ResNet-18 · CIFAR-10 | Median test error, last 10 epochs: 5.17 -> 4.24 | Helped |
| Interpolate raw inputs and labels instead of latent representations at Layer 3 (ablation) | mixup on PreAct ResNet-18 · CIFAR-10 | Median test error, last 10 epochs: 5.39 -> 4.24 | Helped |
| Replace interpolated labels with label smoothing (ablation) | Input-mixing method on PreAct ResNet-18 · CIFAR-10 | Median test error, last 10 epochs: 4.24 -> 5.08 | Hurt |
| Add Gaussian noise instead of mixing labels (ablation) | PreAct ResNet-18 · CIFAR-10 | Median test error, last 10 epochs: 5.53 -> 6.41 | Hurt |
| Increase weight decay from 10⁻⁴ to 5×10⁻⁴ (ablation) | mixup AC + RP on PreAct ResNet-18 · CIFAR-10 | Median test error, last 10 epochs: 4.24 -> 4.68 | Hurt |
Closest papers by meaning
- C-Mixup: Improving Generalization in Regression (NeurIPS 2022 · similarity 0.92)
- Mixup Inference: Better Exploiting Mixup to Defend Adversarial Attacks (ICLR 2020 · similarity 0.91)
- Pushing Boundaries: Mixup's Influence on Neural Collapse (ICLR 2024 · similarity 0.89)
- Adversarial AutoMixup (ICLR 2024 · similarity 0.88)
- Catch-Up Mix: Catch-Up Class for Struggling Filters in CNN (AAAI 2024 · similarity 0.87)
- Selective Mixup Helps with Distribution Shifts, But Not (Only) because of Mixup (ICML 2024 · 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.