Research map / Adversarial robustness and security / Certified and empirical robustness
IPMix: Label-Preserving Data Augmentation Method for Training Robust Classifiers
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 patch-level mixing (ablation) | IPMix · ImageNet, ResNet-50 | mCE: 65.1 -> 63 | Helped |
| add pixel-level mixing (ablation) | IPMix · ImageNet, ResNet-50 | mCE: 65.6 -> 63 | Helped |
| add image-level mixing (ablation) | IPMix · ImageNet, ResNet-50 | mCE: 66.2 -> 63 | Helped |
| add image-level mixing (ablation) | IPMix · ImageNet, ResNet-50 | clean error: 23.5 -> 22.2 | Helped |
| add pixel-level mixing (ablation) | IPMix · CIFAR-100, WRN40-4 | calibration RMS: 8.2 -> 2.8 | Helped |
| add patch-level mixing (ablation) | IPMix · CIFAR-100, WRN40-4 | clean error: 19.7 -> 19.4 | Helped |
| replace fractal mixing set with FractalDB (ablation) | IPMix · CIFAR-100, WRN40-4 | calibration RMS: 2.8 -> 5.4 | Hurt |
| replace fractal mixing set with RCDB (ablation) | IPMix · CIFAR-100, WRN40-4 | mCE: 28.6 -> 28.4 | No clear effect |
| use Fractal + FVis mixing set instead of fractals (ablation) | IPMix · CIFAR-100, WRN40-4 | mCE: 28.6 -> 28.8 | No clear effect |
| restrict mixing operations to addition + multiplication (ablation) | IPMix · CIFAR-100, WRN40-4 | mCE: 28.6 -> 31 | Hurt |
| restrict mixing operations to addition + multiplication (ablation) | IPMix · CIFAR-100, WRN40-4 | clean error: 19.4 -> 19.2 | Helped |
| vary augmentation chains k (2,3,4) and max image augmentation t (2,3) (ablation) | IPMix · CIFAR-100, WRN40-4 | clean error: 19.7 -> 19.4 | No clear effect |
Closest papers by meaning
- DiffuseMix: Label-Preserving Data Augmentation with Diffusion Models (CVPR 2024 · similarity 0.86)
- Toward Robustness in Multi-Label Classification: A Data Augmentation Strategy against Imbalance and Noise (AAAI 2024 · similarity 0.85)
- Adversarial AutoMixup (ICLR 2024 · similarity 0.84)
- DiffAug: A Diffuse-and-Denoise Augmentation for Training Robust Classifiers (NeurIPS 2024 · similarity 0.83)
- PatchDEMUX: A Certifiably Robust Framework for Multi-label Classifiers Against Adversarial Patches (CVPR 2025 · similarity 0.83)
- SAFLEX: Self-Adaptive Augmentation via Feature Label Extrapolation (ICLR 2024 · similarity 0.82)
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.