Research map / Label-efficient and robust learning / Semi-supervised learning
Rethinking the Value of Labels for Improving Class-Imbalanced Learning
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 |
|---|---|---|---|
| add SSL with 5x unlabeled data (pseudo-labeling) | CE · CIFAR-10-LT rho=100 | top-1 error (%): 29.64 -> 17.48 | Helped |
| add SSL with 5x unlabeled data | LDAM-DRW · CIFAR-10-LT rho=100 | top-1 error (%): 22.97 -> 14.96 | Helped |
| add self-supervised pre-training (SSP, Rotation) | CB-CE · CIFAR-10-LT rho=100 | top-1 error (%): 27.63 -> 23.47 | Helped |
| add SSP | LDAM-DRW · CIFAR-100-LT rho=100 | top-1 error (%): 57.96 -> 56.57 | Helped |
| add SSP (MoCo) | cRT ResNet-50 · ImageNet-LT | top-1 error (%): 52.7 -> 48.7 | Helped |
| add SSP (MoCo) | cRT · iNaturalist 2018 | top-1 error (%): 34.8 -> 31.9 | Helped |
| more imbalanced unlabeled data (rho_U=2rho vs 1) (ablation) | CE + D_U@5x · CIFAR-10-LT rho=100 | top-1 error (%): 17.48 -> 20.06 | Hurt |
| Mean Teacher instead of pseudo-label SSL (ablation) | SSL for imbalance · CIFAR-10-LT rho=100 | top-1 error (%): 18.74 -> 16.52 | Helped |
| increase unlabeled data from 0.5x to 10x (ablation) | CE + D_U · CIFAR-10-LT rho=50 | top-1 error (%): 21.75 -> 16.88 | Helped |
| add unlabeled data with low relevance (mismatched classes) (ablation) | SSL for imbalance · CIFAR-10-LT | top-1 error | Hurt |
Closest papers by meaning
- Imbalances in Neurosymbolic Learning: Characterization and Mitigating Strategies (NeurIPS 2025 · similarity 0.90)
- Toward Robustness in Multi-Label Classification: A Data Augmentation Strategy against Imbalance and Noise (AAAI 2024 · similarity 0.89)
- Improved Algorithm for Deep Active Learning under Imbalance via Optimal Separation (ICML 2025 · similarity 0.89)
- Rebalancing Multi-Label Class-Incremental Learning (AAAI 2025 · similarity 0.89)
- Learning Imbalanced Data with Beneficial Label Noise (ICML 2025 · similarity 0.88)
- Robust Visual Recognition with Class-Imbalanced Open-World Noisy Data (AAAI 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.