Research map / Label-efficient and robust learning / Domain adaptation and generalization
Towards Accurate Model Selection in Deep 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 |
|---|---|---|---|
| Use DEV (embedded adapted features with control variate) for model selection instead of source risk | MCD · VisDA | mean accuracy (%): +6.06 over Source Risk (66.36 to 72.42) | Helped |
| Add control variate to DEV (ablation) | DEV on MCD · VisDA | mean accuracy (%): +2.20 (70.22 to 72.42) | Helped |
| Use DEV instead of IWCV for model selection | MCD · VisDA | mean accuracy (%): +5.20 (67.22 to 72.42) | Helped |
| Use DEV instead of source risk for model selection | CDAN · Office-31 avg | mean accuracy (%): +5.17 (82.58 to 87.75) | Helped |
| Use DEV instead of IWCV for model selection | CDAN · Office-31 avg | mean accuracy (%): +4.06 (83.69 to 87.75) | Helped |
| Use DEV instead of source risk for model selection | GTA · Digits avg (USPS/MNIST/SVHN) | mean accuracy (%): +9.04 (85.18 to 94.22) | Helped |
| Use DEV instead of source risk for model selection | PADA · Office-31 avg (partial DA) | mean accuracy (%): +5.92 (87.23 to 93.15) | Helped |
| Use DEV instead of IWCV for model selection | PADA · Office-31 avg (partial DA) | mean accuracy (%): +3.79 (89.36 to 93.15) | Helped |
| Use source risk for model selection in place of target-based selection on SVHN to MNIST | GTA · Digits SVHN->MNIST | accuracy (%): 77.58 vs 93.18 with DEV | Hurt |
| Use IWCV with pre-trained ResNet-50 features (instead of adapted features) for density ratio | MCD · VisDA | mean accuracy (%): 67.22 vs Source Risk 66.36 and DEV 72.42 | Hurt |
| Use DEV's control-variate-free estimator (embedding adapted features) instead of IWCV (ablation) | MCD · VisDA | mean accuracy (%): 70.22 vs 67.22 | Helped |
| Apply DEV to CDAN on Office-31 where source-only selection works well (D->W, W->D) versus large gap (D->A) | CDAN · Office-31 D->A | accuracy (%): Source Risk does well only on small-gap pairs | Mixed |
Closest papers by meaning
- Towards Reliable Model Selection for Unsupervised Domain Adaptation: An Empirical Study and A Certified Baseline (NeurIPS 2024 · similarity 0.93)
- Can We Evaluate Domain Adaptation Models Without Target-Domain Labels? (ICLR 2024 · similarity 0.91)
- Pseudo-Calibration: Improving Predictive Uncertainty Estimation in Unsupervised Domain Adaptation (ICML 2024 · similarity 0.87)
- SimProF: A Simple Probabilistic Framework for Unsupervised Domain Adaptation (AAAI 2025 · similarity 0.87)
- Enhancing Domain Adaptation through Prompt Gradient Alignment (NeurIPS 2024 · similarity 0.87)
- Toward Improving Robustness and Accuracy in Unsupervised Domain Adaptation (AAAI 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.