Research map / Label-efficient and robust learning / Test-time adaptation
Online Adaptation to Label Distribution Shift
Ruihan Wu, Chuan Guo, Yi Su, Kilian Q Weinberger
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 |
|---|---|---|---|
| Use online gradient descent for reweighting instead of the fixed classifier | ResNet-18 · CIFAR-10 simulated label shifts | average error (%) | Improved |
| Use OGD with surrogate-loss gradients instead of finite-difference gradients (ablation) | OGD · CIFAR-10 simulated label shifts | average error (%): 10.71 ± 0.10 -> 10.24 ± 0.06 | Improved |
| Use OGD with finite-difference gradients rather than surrogate-loss gradients (ablation) | OGD · ArXiv test set | average error (%): 25.70 -> 25.52 | Improved |
| Use FTH instead of the optimal fixed classifier | Label-shift adaptation · CIFAR-10 simulated label shifts | average error (%): 10.24 ± 0.08 -> 10.25 ± 0.06 | No clear effect |
| Use OGD instead of the optimal fixed classifier | Label-shift adaptation · CIFAR-10 monotone shift | average error (%): 10.24 ± 0.08 -> 9.75 ± 0.07 | Improved |
| Use FTFWH with window size 100 when the periodicity is 100 (ablation) | FTFWH · CIFAR-10 periodic shift (Tp=100) | average error (%): 10.25 ± 0.08 -> 12.15 ± 0.01 | Hurt |
| Change FTFWH window size from 10^2 to 10^3 (ablation) | FTFWH · ArXiv test set | average error (%): 30.14 -> 26.09 | Improved |
| Use FTFWH instead of FTH or OGD | Online label-shift adaptation · ArXiv test set | average error (%): 25.62 -> 30.14 | Hurt |
| Use OGD instead of the optimal fixed classifier | Label-shift adaptation · ArXiv test set | average error (%): 25.56 -> 25.52 | No clear effect |
Closest papers by meaning
- Adapting to Online Label Shift with Provable Guarantees (NeurIPS 2022 · similarity 0.90)
- Label Shift Meets Online Learning: Ensuring Consistent Adaptation with Universal Dynamic Regret (CVPR 2025 · similarity 0.89)
- Online Feature Updates Improve Online (Generalized) Label Shift Adaptation (NeurIPS 2024 · similarity 0.89)
- Protected Test-Time Adaptation via Online Entropy Matching: A Betting Approach (NeurIPS 2024 · similarity 0.87)
- Adaptive Labeling for Efficient Out-of-distribution Model Evaluation (NeurIPS 2024 · similarity 0.86)
- Learning Label Shift Correction for Test-Agnostic Long-Tailed Recognition (ICML 2024 · 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.
