Research map / Label-efficient and robust learning / Recognition and long-tail
Online Label Shift: Optimal Dynamic Regret meets Practical Algorithms
Dheeraj Baby, Saurabh Garg, Tzu-Ching Yen, Sivaraman Balakrishnan, Zachary Chase Lipton, Yu-Xiang Wang
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 FLH-FTL label marginal estimation instead of FTH | online label shift correction · CIFAR UOLS Bernoulli | classification error (%): 11 -> 10 | Improved |
| use FLH-FTL instead of FTH | online label shift correction · Synthetic UOLS Bernoulli | classification error (%): 6.5 -> 5.4 | Improved |
| use FLH-FTL instead of FTH | online label shift correction · Fashion UOLS Bernoulli | classification error (%): 8.5 -> 7.7 | Improved |
| use FLH-FTL instead of FTH | online label shift correction · MNIST and EuroSAT/ArXiv Sinusoidal shift | classification error (%): 3.3 -> 3.3 | No clear effect |
| use FLH-FTL instead of FTH | online label shift correction · CIFAR UOLS Bernoulli | MSE of label marginal: 0.27 -> 0.17 | Improved |
| use FLH-FTL instead of ROGD | online label shift correction · CIFAR UOLS Bernoulli | MSE of label marginal: 0.39 -> 0.17 | Improved |
| decrease amount of offline holdout data (ablation) | FLH-FTL vs alternatives · CIFAR10 UOLS monotone shift | classification error and marginal MSE | No clear effect |
| increase offline training data at expense of holdout (ablation) | FLH-FTL · CIFAR10 UOLS monotone shift | classification error | Improved |
| continual training with re-sampling (CT-RS) instead of naive continual training | SOLS baseline · CIFAR10/MNIST SOLS Bernoulli | classification performance | Improved |
| CT-RS instead of weighted ERM from scratch | SOLS · MNIST/CIFAR SOLS | computation cost: 5--15x faster | Improved |
| FLH-FTL instead of FTH for marginal estimation in SOLS | CT-RS · MNIST/CIFAR SOLS | marginal estimation and classification | Improved |
| use FLH-FTL instead of ATLAS/UOGD | online label shift correction · CIFAR UOLS Bernoulli | classification error (%): 13 -> 10 | Improved |
Closest papers by meaning
- Label Shift Meets Online Learning: Ensuring Consistent Adaptation with Universal Dynamic Regret (CVPR 2025 · similarity 0.94)
- Efficient Non-stationary Online Learning by Wavelets with Applications to Online Distribution Shift Adaptation (ICML 2024 · similarity 0.89)
- Online Feature Updates Improve Online (Generalized) Label Shift Adaptation (NeurIPS 2024 · similarity 0.88)
- Stochastic Online Conformal Prediction with Semi-Bandit Feedback (ICML 2025 · similarity 0.87)
- Non-stationary Online Learning for Curved Losses: Improved Dynamic Regret via Mixability (ICML 2025 · similarity 0.86)
- Online Classification with Predictions (NeurIPS 2024 · similarity 0.86)
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.
