Research map / Label-efficient and robust learning / Domain adaptation and generalization
SRoUDA: Meta Self-Training for Robust Unsupervised Domain Adaptation
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 |
|---|---|---|---|
| Add meta self-training with adversarial target-model training and source-model meta-updates | UDA model · Office-31 average | robust accuracy (PGD-20): 0.43 to 75.79 | Helped |
| Use SRoUDA instead of the UDA baseline | UDA model · Digits average | robust accuracy (PGD-20): 29.58 % to 90.40 % | Helped |
| Use SRoUDA instead of the UDA baseline | UDA model · Office-31, A→W | clean accuracy: 94.50 % to 95.97 % | Helped |
| Use SRoUDA instead of naive self-training with AT | UDA model · Office-31 average | clean accuracy: 79.78 % to 87.27 | Helped |
| Add source-model pre-training (ablation) | SRoUDA · W→A | clean accuracy: 50.75 to 67.10 | Helped |
| Add source-model pre-training (ablation) | SRoUDA · W→A | robust accuracy (PGD-20): 40.22 to 57.42 | Helped |
| Add source-model pre-training (ablation) | SRoUDA · D→A | clean accuracy: 40.43 to 72.47 | Helped |
| Add source-model pre-training (ablation) | SRoUDA · D→A | robust accuracy (PGD-20): 22.15 to 57.20 | Helped |
| Add the meta-step to fine-tune the source model (ablation) | SRoUDA · W→A | clean accuracy: 64.52 to 67.10 | Helped |
| Add the meta-step to fine-tune the source model (ablation) | SRoUDA · W→A | robust accuracy (PGD-20): 53.12 to 57.42 | Helped |
| Add the meta-step to fine-tune the source model (ablation) | SRoUDA · D→A | clean accuracy: 68.17 to 72.47 | Helped |
| Add the meta-step to fine-tune the source model (ablation) | SRoUDA · D→A | robust accuracy (PGD-20): 54.84 to 57.20 | Helped |
Closest papers by meaning
- Toward Improving Robustness and Accuracy in Unsupervised Domain Adaptation (AAAI 2025 · similarity 0.92)
- Can We Evaluate Domain Adaptation Models Without Target-Domain Labels? (ICLR 2024 · similarity 0.89)
- Adversarial Self-Training Improves Robustness and Generalization for Gradual Domain Adaptation (NeurIPS 2023 · similarity 0.88)
- Guiding Pseudo-Labels With Uncertainty Estimation for Source-Free Unsupervised Domain Adaptation (CVPR 2023 · similarity 0.87)
- 3SAT: A Simple Self-Supervised Adversarial Training Framework (AAAI 2025 · similarity 0.87)
- CASUAL: Conditional Support Alignment for Domain Adaptation with Label Shift (AAAI 2025 · similarity 0.85)
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.