Research map / Label-efficient and robust learning / Domain adaptation and generalization
StyleAdv: Meta Style Adversarial Training for Cross-Domain Few-Shot 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 |
|---|---|---|---|
| Train with adversarially perturbed styles rather than the original StyleAdv baseline | ResNet-10 + GNN · Eight target datasets, 5-way 1-shot | accuracy (%): 43.55 -> 48.06 | Helped |
| Train with adversarially perturbed styles rather than the original StyleAdv baseline | ResNet-10 + GNN · Eight target datasets, 5-way 5-shot | accuracy (%): 58.84 -> 63.77 | Helped |
| Add target-example finetuning | StyleAdv with ResNet-10 + GNN · Eight target datasets, 5-way 5-shot | accuracy (%): 63.77 -> 67.28 | Helped |
| Attack styles instead of image pixels (ablation) | StyleAdv training, ResNet-10 + GNN · Eight target datasets, 5-way 1-shot | accuracy (%): 46.96 -> 48.06 | Helped |
| Attack styles instead of features (ablation) | StyleAdv training, ResNet-10 + GNN · Eight target datasets, 5-way 5-shot | accuracy (%): 62.08 -> 63.77 | Helped |
| Use one-step Style-FGSM instead of multi-step Style-PGD (ablation) | StyleAdv with ResNet-10 + GNN · Eight target datasets, 5-way 1-shot | accuracy (%): 46.97 -> 48.06 | Helped |
| Add the global classification loss L_cls (ablation) | StyleAdv · Eight target datasets, 5-way 1-shot | accuracy (%): 46.14 -> 48.06 | Helped |
| Add the consistency loss L_cons (ablation) | StyleAdv · Eight target datasets, 5-way 1-shot | accuracy (%): 47.24 -> 48.06 | Helped |
| Add the adversarial FSL loss L_fsl^adv and consistency loss L_cons (ablation) | StyleAdv · Eight target datasets, 5-way 1-shot | accuracy (%): 45.75 -> 48.06 | Helped |
| Add the original FSL loss L_fsl and consistency loss L_cons (ablation) | StyleAdv · Eight target datasets, 5-way 1-shot | accuracy (%): 46.84 -> 48.06 | Helped |
| Use a random start in the style attack (ablation) | StyleAdv · Eight target datasets, 5-way 1-shot | accuracy (%): 47.64 -> 48.06 | Helped |
| Set p_skip to 0.4 instead of 0 (ablation) | StyleAdv · Eight target datasets, 5-way 1-shot | accuracy (%): 47.34 -> 48.06 | Helped |
Closest papers by meaning
- SVasP: Self-Versatility Adversarial Style Perturbation for Cross-Domain Few-Shot Learning (AAAI 2025 · similarity 0.92)
- Pushing the Limit of Fine-Tuning for Few-Shot Learning: Where Feature Reusing Meets Cross-Scale Attention (AAAI 2024 · similarity 0.86)
- Style Adaptation and Uncertainty Estimation for Multi-Source Blended-Target Domain Adaptation (NeurIPS 2024 · similarity 0.86)
- Flatten Long-Range Loss Landscapes for Cross-Domain Few-Shot Learning (CVPR 2024 · similarity 0.85)
- Few-Shot Adversarial Prompt Learning on Vision-Language Models (NeurIPS 2024 · similarity 0.85)
- StyDeSty: Min-Max Stylization and Destylization for Single Domain Generalization (ICML 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.