Research map / Label-efficient and robust learning / Semi-supervised learning
CReST: A Class-Rebalancing Self-Training Framework for Imbalanced Semi-Supervised Learning
Check 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 class-rebalancing self-training | FixMatch · CIFAR10-LT, β=10%, γ=100 | accuracy (%): 66.3 -> 75.9 | Helped |
| Add progressive distribution alignment to CReST | FixMatch · CIFAR10-LT, β=10%, γ=100 | accuracy (%): 75.9 -> 78.1 | Helped |
| Add class-rebalancing self-training | FixMatch · CIFAR100-LT, β=10%, γ=100 | accuracy (%): 28.3 -> 32.1 | Helped |
| Add progressive distribution alignment to CReST | FixMatch · CIFAR100-LT, β=10%, γ=100 | accuracy (%): 32.1 -> 34.6 | Helped |
| Add class-rebalancing self-training, progressive distribution alignment, and logit adjustment | FixMatch · CIFAR10-LT, β=10%, γ=100 | accuracy (%): 67.3 -> 81.2 | Helped |
| Replace DARP with CReST+ | FixMatch · CIFAR10, γ=100 (DARP protocol) | accuracy (%): 75.5 -> 77.4 | Helped |
| Use CReST+ across three self-training generations | FixMatch · ImageNet127, 10% labels | average class recall: 65.8 -> 73.7 | Helped |
| Set α=0, removing class-rebalancing sampling and adding all pseudo-labels (ablation) | CReST · CIFAR10-LT, β=10%, γ=100 | accuracy | No clear effect |
| Increase α to 1, strengthening minority-class sampling (ablation) | CReST · CIFAR10-LT, β=10%, γ=100 | accuracy: 3.0% drop | Hurt |
| Use constant distribution-alignment temperature t=0.5 instead of the original t=1.0 (ablation) | FixMatch with distribution alignment · CIFAR10-LT, single generation | accuracy: 70.0% -> 75.1% | Helped |
| Decrease the constant temperature to t=0.1 (ablation) | FixMatch with distribution alignment · CIFAR10-LT, single generation | accuracy | Hurt |
| Use a constant low temperature rather than progressively lowering temperature across generations (ablation) | CReST+ · CIFAR10-LT, multiple generations | accuracy | Hurt |
Closest papers by meaning
- Twice Class Bias Correction for Imbalanced Semi-supervised Learning (AAAI 2024 · similarity 0.90)
- BaCon: Boosting Imbalanced Semi-supervised Learning via Balanced Feature-Level Contrastive Learning (AAAI 2024 · similarity 0.90)
- CDMAD: Class-Distribution-Mismatch-Aware Debiasing for Class-Imbalanced Semi-Supervised Learning (CVPR 2024 · similarity 0.89)
- LCGC: Learning from Consistency Gradient Conflicting for Class-Imbalanced Semi-Supervised Debiasing (AAAI 2025 · similarity 0.89)
- BEM: Balanced and Entropy-based Mix for Long-Tailed Semi-Supervised Learning (CVPR 2024 · similarity 0.87)
- Learning under Localized Minority Imbalance (NeurIPS 2026 · similarity 0.87)
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.