Research map / Adversarial robustness and security / Certified and empirical robustness
Puzzle Mix: Exploiting Saliency and Local Statistics for Optimal Mixup
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 saliency-based multi-label masking and transport instead of Manifold mixup | WRN28-10 · CIFAR-100 | Top-1 Error(%): 17.40 -> 15.95 | Helped |
| Add stochastic adversarial training to Puzzle Mix | WRN28-10 · CIFAR-100 | FGSM Error(%): 63.71 -> 47.18 | Helped |
| Use saliency-based multi-label masking and transport instead of Manifold mixup | PreActResNet18 · CIFAR-100 | Top-1 Error(%): 20.98 -> 19.62 | Helped |
| Add stochastic adversarial training to Puzzle Mix | PreActResNet18 · CIFAR-100 | FGSM Error(%): 70.09 -> 57.11 | Helped |
| Use saliency-based multi-label masking and transport instead of Manifold mixup | PreActResNet18 · Tiny-ImageNet | Top-1 Error(%): 41.99 -> 36.52 | Helped |
| Add stochastic adversarial training to Puzzle Mix | PreActResNet18 · Tiny-ImageNet | FGSM Error(%): 87.19 -> 82.07 | Helped |
| Use Puzzle Mix rather than CutMix | ResNet-50 · ImageNet, 100 epochs | Top-1 Error(%): 22.92 -> 22.49 | Helped |
| Use Puzzle Mix on two AugMix images rather than the AugMix JSD strategy | WRN28-10 · CIFAR-100 | Top-1 Error(%): 20.45 -> 16.50 | Helped |
| Use Puzzle Mix on two AugMix images rather than the AugMix JSD strategy | WRN28-10 · CIFAR-100-C | Corruption Error(%): 32.22 -> 29.91 | Helped |
| Use binary rather than multi-label masks (ablation) | Puzzle Mix, WRN28-10 · CIFAR-100 | Top-1 Error(%): 15.95 -> 17.39 | Hurt |
| Remove transport (ablation) | Puzzle Mix, WRN28-10 · CIFAR-100 | Top-1 Error(%): 15.95 -> 16.38 | Hurt |
| Remove linear decay of adversarial perturbation (ablation) | Puzzle Mix adversarial training, WRN28-10 · CIFAR-100 | FGSM Error(%): 47.18 -> 43.16 | Helped |
Closest papers by meaning
- mixup: Beyond Empirical Risk Minimization (ICLR 2018 · similarity 0.90)
- Saliency Grafting: Innocuous Attribution-Guided Mixup with Calibrated Label Mixing (AAAI 2022 · similarity 0.88)
- Co-Mixup: Saliency Guided Joint Mixup with Supermodular Diversity (ICLR 2021 · similarity 0.88)
- Catch-Up Mix: Catch-Up Class for Struggling Filters in CNN (AAAI 2024 · similarity 0.86)
- Adversarial AutoMixup (ICLR 2024 · similarity 0.86)
- Pushing Boundaries: Mixup's Influence on Neural Collapse (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.