Research map / Label-efficient and robust learning / Domain adaptation and generalization
Generalizing to Unseen Domains via Adversarial Data Augmentation
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 adversarial data augmentation instead of ERM | ConvNet trained on MNIST · SVHN | accuracy: $0.283 \pm 0.032$ -> $0.306 \pm 0.008$ | Helped |
| Use adversarial data augmentation instead of ERM | ConvNet trained on MNIST · MNIST-M | accuracy: $0.548 \pm 0.021$ -> $0.579 \pm 0.009$ | Helped |
| Use adversarial data augmentation instead of ERM | ConvNet trained on MNIST · SYN | accuracy: $0.406 \pm 0.022$ -> $0.431 \pm 0.017$ | Helped |
| Use adversarial data augmentation instead of ERM | ConvNet trained on MNIST · USPS | accuracy: $0.789 \pm 0.017$ -> $0.772 \pm 0.010$ | Hurt |
| Constrain adversarial perturbations in semantic space rather than pixel space (ablation) | Adversarial Data Augmentation · SVHN, MNIST-M, SYN, and USPS | accuracy | Helped |
| Increase the number of augmentation iterations (ablation) | Adversarial Data Augmentation · SVHN, MNIST-M, and SYN | accuracy | Helped |
| Add adversarial data augmentation to ridge regularization (ablation) | ConvNet trained on MNIST · SVHN, MNIST-M, SYN, and USPS | accuracy | Helped |
| Choose among models trained with different γ values using maximum softmax confidence (ablation) | Adversarial Data Augmentation ensemble · Digit target domains | accuracy | Helped |
| Use adversarial data augmentation instead of ERM | FCN with ResNet-50 trained on SYTHIA · Cross-domain segmentation, especially models trained on Night scenarios | mIoU | Helped |
| Use adversarial data augmentation instead of ERM | FCN with ResNet-50 trained on NY.Like C./Spring · Highway/Dawn | mIoU: $\mathbf{17.1}$ -> $14.5$ | Hurt |
Closest papers by meaning
- Learning to Learn Single Domain Generalization (CVPR 2020 · similarity 0.91)
- Domain Generalization With Adversarial Feature Learning (CVPR 2018 · similarity 0.91)
- Adversarially Adaptive Normalization for Single Domain Generalization (CVPR 2021 · similarity 0.88)
- Adversarial Dropout Regularization (ICLR 2018 · similarity 0.88)
- AdvST: Revisiting Data Augmentations for Single Domain Generalization (AAAI 2024 · similarity 0.87)
- Sharpness-Aware Model-Agnostic Long-Tailed Domain Generalization (AAAI 2024 · similarity 0.87)
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.