Research map / Label-efficient and robust learning / Domain adaptation and generalization
Automated Synthetic-to-Real Generalization
Wuyang Chen, Zhiding Yu, Zhangyang Wang, Animashree Anandkumar
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 ImageNet prediction-matching Proxy Guidance to large learning rates for all layers (ablation) | ResNet-101 · VisDA-17 classification | Top-1 accuracy (%): 28.2 -> 58.7 (+30.5) | Improved |
| Add ImageNet prediction-matching Proxy Guidance to small-backbone/large-FC learning rates (ablation) | ResNet-101 · VisDA-17 classification | Top-1 accuracy (%): 49.3 -> 60.2 (+10.9) | Improved |
| Increase Proxy Guidance strength from lambda 0.01 to 0.1 (ablation) | ResNet-101 · VisDA-17 classification | Accuracy (%): 58.9 -> 60.1 | Improved |
| Increase Proxy Guidance strength from lambda 0.1 to 0.5 (ablation) | ResNet-101 · VisDA-17 classification | Accuracy (%): 60.1 -> 58.5 | Hurt |
| Use Proxy Guidance instead of naive synthetic training | FCN-Res50 · GTA5 → Cityscapes | mIoU (%): 23.29 -> 31.89 | Improved |
| Use Proxy Guidance instead of naive synthetic training | FCN-VGG16 · GTA5 → Cityscapes | mIoU (%): 19.89 -> 31.47 | Improved |
| Initialize CRST (MRKLD) with the ASG source model | ResNet-101 · VisDA-17 adaptation | Top-1 accuracy (%): 77.9 (0.5) -> 84.6 (0.4) | Improved |
| Use RL-L2O rather than hand-tuned learning-rate schedules (ablation) | ASG · VisDA-17 classification | Generalization accuracy | Improved |
| Use Proxy Guidance rather than naively freezing the backbone (ablation) | ResNet-101 · VisDA-17 classification | Top-1 accuracy (%): 53.3 -> 60.2 | Improved |
Closest papers by meaning
- Real-Fake: Effective Training Data Synthesis Through Distribution Matching (ICLR 2024 · similarity 0.89)
- Provably Improving Generalization of Few-shot models with Synthetic Data (ICML 2025 · similarity 0.89)
- The Unmet Promise of Synthetic Training Images: Using Retrieved Real Images Performs Better (NeurIPS 2024 · similarity 0.89)
- On the Stability of Iterative Retraining of Generative Models on their own Data (ICLR 2024 · similarity 0.87)
- Scaling Laws of Synthetic Images for Model Training ... for Now (CVPR 2024 · similarity 0.87)
- Improving the Scaling Laws of Synthetic Data with Deliberate Practice (ICML 2025 · 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.
