Research map / Label-efficient and robust learning / Recognition and long-tail
Implicit Semantic Data Augmentation for Deep Networks
Yulin Wang, Xuran Pan, Shiji Song, Hong Zhang, Gao Huang, Cheng Wu
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 |
|---|---|---|---|
| Add ISDA | ResNet-50 · ImageNet val | Top-1 error rate: 23.0 -> 21.9 | Improved |
| Add ISDA | ResNet-101 · ImageNet val | Top-1 error rate: 21.7 -> 20.8 | Improved |
| Add ISDA | DenseNet-BC-121 · ImageNet val | Top-1 error rate: 23.7 -> 23.2 | Improved |
| Add ISDA | ResNet-110 · CIFAR-10 | error rate: 6.76 ± 0.34% -> 6.33 ± 0.19% | Improved |
| Add ISDA | ResNet-110 · CIFAR-100 | error rate: 28.67 ± 0.44% -> 27.57 ± 0.46% | Improved |
| Add ISDA to AutoAugment | Wide-ResNet-28-10 · CIFAR-100 | error rate: 16.60 ± 0.40% -> 15.62 ± 0.32% | Improved |
| Add ISDA to Cutout | Wide-ResNet-28-10 · CIFAR-100 | error rate: 18.05 ± 0.25% -> 17.02 ± 0.11% | Improved |
| Use class-conditional covariance rather than identity covariance (ablation) | ISDA · CIFAR-10 | error rate: 3.63±0.12% -> 3.58±0.15% | Improved |
| Use class-conditional covariance rather than identity covariance (ablation) | ISDA · CIFAR-100 | error rate: 18.53±0.02% -> 17.98±0.15% | Improved |
| Retain covariance correlations instead of using only diagonal entries (ablation) | ISDA · CIFAR-10 | error rate: 3.70±0.15% -> 3.58±0.15% | Improved |
| Retain covariance correlations instead of using only diagonal entries (ablation) | ISDA · CIFAR-100 | error rate: 18.23±0.02% -> 17.98±0.15% | Improved |
| Use class-specific covariance matrices rather than one global covariance (ablation) | ISDA · CIFAR-10 | error rate: 3.67±0.07% -> 3.58±0.15% | Improved |
Closest papers by meaning
- Regularizing Deep Networks with Semantic Data Augmentation (TPAMI 2021 · similarity 0.96)
- SF(DA)$^2$: Source-free Domain Adaptation Through the Lens of Data Augmentation (ICLR 2024 · similarity 0.85)
- Domain Gap Embeddings for Generative Dataset Augmentation (CVPR 2024 · similarity 0.85)
- AdvST: Revisiting Data Augmentations for Single Domain Generalization (AAAI 2024 · similarity 0.84)
- First-Order Manifold Data Augmentation for Regression Learning (ICML 2024 · similarity 0.84)
- LLM-AutoDA: Large Language Model-Driven Automatic Data Augmentation for Long-tailed Problems (NeurIPS 2024 · similarity 0.84)
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.
