Research map / Adversarial robustness and security / Certified and empirical robustness
IPF-RDA: An Information-Preserving Framework for Robust Data Augmentation
Suorong Yang, Hongchao Yang, Suhan Guo, Furao Shen, Jian Zhao
Read the paperCheck your idea against 137,217 papers
AI-extracted. The measured effects below were read by AI from the paper's full text (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 |
|---|---|---|---|
| integrate Cutout into IPF | Cutout · CIFAR-10 test (ResNet50) | error rate: 4.19 -> 3.27 | Improved |
| integrate RandAugment into IPF | RandAugment · CIFAR-10 test (ResNet50) | error rate: 3.75 -> 2.82 | Improved |
| integrate TrivialAugment into IPF | TrivialAugment · CIFAR-10 test (ResNet50) | error rate: 2.87 -> 2.85 | No clear effect |
| integrate RandomErasing into IPF | RandomErasing · Tiny-ImageNet val (ResNet-18) | error rate: 36.00 -> 28.58 | Improved |
| integrate Cutout into IPF on DenseNet201 | Cutout · Tiny-ImageNet val (DenseNet201) | error rate: 30.68 -> 30.60 | No clear effect |
| integrate HaS into IPF for ReID | HaS · Market1501 ReID | mAP: 86.9 -> 88.4 | Improved |
| integrate HaS into IPF in semi-supervised UDA | HaS · CIFAR-10 semi-supervised (4000 labels) | accuracy: 86.96 -> 91.32 | Improved |
| integrate FixMatch into IPF with 400 labels | FixMatch · CIFAR-100 semi-supervised (400 labels) | error rate: 48.85 -> 45.17 | Improved |
| add rectangular region preservation instead of point-level preservation (ablation) | IPF-TrivialAugment · CIFAR-10 test (ResNet-18) | error rate: 5.60 -> 3.18 | Improved |
| use IPF-Cutout instead of KeepAugment saliency-based preservation | Cutout · CIFAR-10 test (ResNet-18, length 16) | error rate: 3.9 -> 3.58 | Improved |
| use IPF instead of Co-Mixup saliency mixing | Co-Mixup · Tiny-ImageNet (PreActR18) | error rate: 35.85 -> 34.33 | Improved |
| raise preservation threshold tau from 20% to 60% (l=16) (ablation) | IPF-Cutout · CIFAR-10 test | error rate: 3.67 -> 3.57 | No clear effect |
Closest papers by meaning
- IPMix: Label-Preserving Data Augmentation Method for Training Robust Classifiers (NeurIPS 2023 · similarity 0.87)
- AugMax: Adversarial Composition of Random Augmentations for Robust Training (NeurIPS 2021 · similarity 0.86)
- Regularizing Deep Networks with Semantic Data Augmentation (TPAMI 2021 · similarity 0.85)
- Tradeoffs in Data Augmentation: An Empirical Study (ICLR 2021 · similarity 0.84)
- When Dynamic Data Selection Meets Data Augmentation: Achieving Enhanced Training Acceleration (ICML 2025 · similarity 0.84)
- A CLIP-Powered Framework for Robust and Generalizable Data Selection (ICLR 2025 · similarity 0.83)
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.
