Research map / Label-efficient and robust learning / Zero-/few-shot and meta-learning
Data Augmentation for Meta-Learning
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 |
|---|---|---|---|
| Apply CutMix to query images rather than using baseline augmentation (ablation) | R2-D2 with ResNet-12 · CIFAR-FS 1-shot | accuracy: 71.95 ± 0.37 -> 75.97 ± 0.34 | Helped |
| Apply CutMix to support images rather than using baseline augmentation (ablation) | R2-D2 with ResNet-12 · CIFAR-FS 1-shot | accuracy: 71.95 ± 0.37 -> 72.79 ± 0.37 | No clear effect |
| Add support CutMix to query CutMix (ablation) | R2-D2 with ResNet-12 · CIFAR-FS 1-shot | accuracy: 75.97 ± 0.34 -> 75.00 ± 0.37 | Hurt |
| Add task rotation to query CutMix (ablation) | R2-D2 with ResNet-12 · CIFAR-FS 5-shot | accuracy: 87.28 ± 0.23 -> 87.68 ± 0.24 | Helped |
| Apply random crop as shot augmentation (ablation) | R2-D2 with ResNet-12 · CIFAR-FS 1-shot | accuracy: 71.95 ± 0.37 -> 70.56 ± 0.37 | Hurt |
| Restrict the query pool to 5 samples per class (ablation) | R2-D2 with ResNet-12 · CIFAR-FS 1-shot | accuracy: 71.73 ± 0.37 -> 60.25 ± 0.37 | Hurt |
| Restrict the support pool to 5 samples per class (ablation) | R2-D2 with ResNet-12 · CIFAR-FS 1-shot | accuracy: 71.73 ± 0.37 -> 70.97 ± 0.36 | No clear effect |
| Restrict the task pool to 13 tasks (ablation) | R2-D2 with ResNet-12 · CIFAR-FS 1-shot | accuracy: 71.73 ± 0.37 -> 68.24 ± 0.38 | Hurt |
| Use Meta-MaxUp instead of baseline augmentation | R2-D2 with ResNet-12 · CIFAR-FS 5-shot | accuracy: 84.56 ± 0.25 -> 88.57 ± 0.24 | Helped |
| Train with Meta-MaxUp | MetaOptNet with ResNet-12 · Meta-Dataset Omniglot | accuracy: 97.43 ± 0.10 -> 96.59 ± 0.09 | Hurt |
| Train with Meta-MaxUp | MetaOptNet with ResNet-12 · Meta-Dataset VGG Flower | accuracy: 88.16 ± 0.25 -> 91.13 ± 0.15 | Helped |
Closest papers by meaning
- MetaAugment: Sample-Aware Data Augmentation Policy Learning (AAAI 2021 · similarity 0.90)
- Meta-Dataset: A Dataset of Datasets for Learning to Learn from Few Examples (ICLR 2020 · similarity 0.89)
- On sensitivity of meta-learning to support data (NeurIPS 2021 · similarity 0.88)
- Meta-Learning without Memorization (ICLR 2020 · similarity 0.88)
- Improving Generalization via Meta-Learning on Hard Samples (CVPR 2024 · similarity 0.87)
- Effective Data Augmentation With Diffusion Models (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.