Research map / Label-efficient and robust learning / Domain adaptation and generalization
Cycle Self-Training for 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 |
|---|---|---|---|
| Use cycle self-training instead of the strongest listed feature-adaptation or self-training baseline | CST · Office-Home (Avg.) | accuracy: 72.2 -> 73.0 | Helped |
| Use cycle self-training instead of VAT+Entropy | CST · Multi-Domain Sentiment Dataset (Avg.) | accuracy: 90.1 -> 91.5 | Helped |
| Use cycle self-training instead of MDD+FixMatch | VisDA-2017, ResNet-50 · VisDA-2017 | mean class accuracy: 77.8 ± 0.3 -> 79.9 ± 0.5 | Helped |
| Use cycle self-training instead of STAR | VisDA-2017, ResNet-101 · VisDA-2017 | mean class accuracy: 82.7 -> 84.8 ± 0.6 | Helped |
| Add sharpness-aware regularization | CST on VisDA-2017 · VisDA-2017, ResNet-50 | mean class accuracy: 79.9 ± 0.5 -> 80.6 ± 0.5 | Helped |
| Add sharpness-aware regularization | CST on VisDA-2017 · VisDA-2017, ResNet-101 | mean class accuracy: 84.8 ± 0.6 -> 86.5 ± 0.7 | Helped |
| Add Tsallis entropy regularization (ablation) | CST · VisDA-2017 | accuracy: 72.0 ± 0.4 -> 79.9 ± 0.5 | Helped |
| Replace standard entropy with Tsallis entropy (ablation) | CST · VisDA-2017 | accuracy: 76.2 ± 0.6 -> 79.9 ± 0.5 | Helped |
| Add Tsallis entropy to standard self-training | FixMatch · VisDA-2017 | accuracy: 74.5 ± 0.2 -> 76.3 ± 0.8 | Helped |
| Use CST instead of standard self-training | VisDA-2017 · VisDA-2017 | accuracy: CST performs 5.4% better than FixMatch | Helped |
| Use CST instead of FixMatch with Tsallis entropy | VisDA-2017 · VisDA-2017 | accuracy: FixMatch+Tsallis is 3.6% behind CST | Helped |
| Use CST instead of FixMatch with Tsallis entropy | VisDA-2017 · VisDA-2017 | total variation distance: 0.15 -> 0.12 | Helped |
Closest papers by meaning
- C-SFDA: A Curriculum Learning Aided Self-Training Framework for Efficient Source Free Domain Adaptation (CVPR 2023 · similarity 0.88)
- A Free Lunch for Unsupervised Domain Adaptive Object Detection without Source Data (AAAI 2021 · similarity 0.88)
- Theoretical Analysis of Self-Training with Deep Networks on Unlabeled Data (ICLR 2021 · similarity 0.88)
- Guiding Pseudo-Labels With Uncertainty Estimation for Source-Free Unsupervised Domain Adaptation (CVPR 2023 · similarity 0.88)
- Toward Improving Robustness and Accuracy in Unsupervised Domain Adaptation (AAAI 2025 · similarity 0.87)
- Can We Evaluate Domain Adaptation Models Without Target-Domain Labels? (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.