Research map / Label-efficient and robust learning / Domain adaptation and generalization
Larger Norm More Transferable: An Adaptive Feature Norm Approach for Unsupervised Domain Adaptation
Ruijia Xu, Guanbin Li, Jihan Yang, Liang Lin
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 stepwise feature norm enlargement (SAFN) instead of hard norm target (HAFN) | AFN · Office-Home (vanilla) | accuracy (%): 65.4 -> 67.3 | Improved |
| use SAFN instead of HAFN | AFN · VisDA2017 (vanilla) | per-class accuracy (%): 73.9 -> 76.1 | Improved |
| use SAFN instead of HAFN | AFN · Office-31 | accuracy (%): 83.9 -> 85.7 | Improved |
| use SAFN instead of HAFN | AFN · Office-Home (partial) | accuracy (%): 67.51 -> 71.83 | Improved |
| add entropy minimization (ENT) to SAFN (ablation) | SAFN · ImageCLEF-DA | accuracy (%): 88.1 -> 88.9 | Improved |
| add entropy minimization (ENT) to SAFN (ablation) | SAFN · Office-31 | accuracy (%): 85.7 -> 87.1 | Improved |
| increase unlabeled target data from 25% to 100% (ablation) | SAFN · VisDA2017 | target accuracy | Improved |
| increase feature-norm limit R beyond optimum (ablation) | HAFN · VisDA2017 and Office-31 A to W | accuracy | Mixed |
| vary step size delta r (0.2-0.5 on VisDA, 0.5-2.0 on A to W) (ablation) | SAFN · VisDA2017 and Office-31 A to W | accuracy | No clear effect |
| vary embedding size from 500 to 2000 (ablation) | SAFN · VisDA2017 and Office-31 A to W | accuracy | No clear effect |
| use SAFN instead of adversarial DANN under partial setting | SAFN · Office-Home (partial) | accuracy (%): 47.39 -> 71.83 | Improved |
| use SAFN instead of best baseline on robustness metrics | SAFN · Ar to Rw (5%) partial | PNG (lower is better): SAFN -9.6 vs DAN -0.8 and DANN 4.7 | Improved |
Closest papers by meaning
- Quantifying and Improving Transferability in Domain Generalization (NeurIPS 2021 · similarity 0.90)
- On Learning Domain-Invariant Representations for Transfer Learning with Multiple Sources (NeurIPS 2021 · similarity 0.90)
- Min-Max Statistical Alignment for Transfer Learning (CVPR 2019 · similarity 0.89)
- Learning to Transfer with von Neumann Conditional Divergence (AAAI 2022 · similarity 0.89)
- Domain Adaptation with Conditional Transferable Components (ICML 2016 · similarity 0.89)
- Transferring Knowledge From Large Foundation Models to Small Downstream Models (ICML 2024 · similarity 0.88)
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.
