Research map / Label-efficient and robust learning / Domain adaptation and generalization
When Invariant Representation Learning Meets Label Shift: Insufficiency and Theoretical Insights
You-Wei Luo, Chuan-Xian Ren
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 label-shift weighting (GLS) to conditional shift correction (ablation) | KECA · Office-Home subsampled source (avg over rates) | accuracy (%): 69.9 -> 71.4 | Improved |
| add label-shift weighting at 10% subsample rate (ablation) | KECA · Office-Home subsampled target | accuracy (%): 60.6 -> 65.4 | Improved |
| add label-shift weighting at 10% subsample rate (ablation) | KECA · Office-Home subsampled source | accuracy (%): 65.4 -> 66.5 | Improved |
| add conditional matching to marginal-only corrections (ablation) | label/covariate shift models · Office-Home subsampled source | accuracy (%): 57.5 -> 69.9 | Improved |
| use label shift model instead of covariate shift model under severe shift (ablation) | marginal alignment · Office-Home subsampled source, 10% rate | accuracy (%): 55.3 -> 56.8 | Improved |
| use Gaussian instead of linear kernel (ablation) | KECA · Office-31 subsampled | accuracy (%): 85.2 -> 86.7 | Improved |
| use Gaussian instead of polynomial kernel (ablation) | KECA · Office-31 subsampled | accuracy (%): 86.7 -> 86.7 | No clear effect |
| use Gaussian instead of Laplacian kernel (ablation) | KECA · Office-31 subsampled | accuracy (%): 85.9 -> 86.7 | Improved |
| add importance weighting to DANN | DANN · Office-Home subsampled | accuracy (%): 51.8 -> 57.6 | Improved |
| add importance weighting to CDAN | CDAN · Office-Home subsampled | accuracy (%): 56.3 -> 61.2 | Improved |
| add importance weighting to JAN | JAN · Office-Home subsampled | accuracy (%): 53.9 -> 55.9 | Improved |
| use KECA instead of OSTAR | KECA · Office-Home subsampled | accuracy (%): 59.5 -> 65.9 | Improved |
Closest papers by meaning
- Heterogeneous Label Shift: Theory and Algorithm (ICML 2025 · similarity 0.87)
- Learning Label Shift Correction for Test-Agnostic Long-Tailed Recognition (ICML 2024 · similarity 0.87)
- Classification under Nuisance Parameters and Generalized Label Shift in Likelihood-Free Inference (ICML 2024 · similarity 0.86)
- Generalization Analysis for Label-Specific Representation Learning (NeurIPS 2024 · similarity 0.85)
- Label Shift Meets Online Learning: Ensuring Consistent Adaptation with Universal Dynamic Regret (CVPR 2025 · similarity 0.85)
- Sufficient Invariant Learning for Distribution Shift (CVPR 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.
