Research map / Label-efficient and robust learning / Semi-supervised learning
Don’t fear the unlabelled: safe semi-supervised learning via debiasing
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 debiasing term (labelled H subtracted, unlabelled H kept) to Fixmatch | Fixmatch · CIFAR-10 (n_l=4000) | Accuracy: +1.57% over Fixmatch | Helped |
| Add debiasing term to Fixmatch | Fixmatch · CIFAR-10 (n_l=4000) | Worst class accuracy: +4.91% on worst class | Helped |
| Add debiasing term to Fixmatch | Fixmatch · CIFAR-10 (n_l=4000) and CIFAR-100 (n_l=10000) | Cross entropy: CIFAR-10 0.27 to 0.20; CIFAR-100 1.52 to 1.42 | Helped |
| Add debiasing term to Fixmatch | Fixmatch · CIFAR-100 (n_l=10000) | Accuracy: 69.28 to 71.22 | Helped |
| Add debiasing term to Pseudo-label on MNIST | Pseudo-label · MNIST (n_l=1000) | Accuracy | No clear effect |
| Add debiasing term to Pseudo-label on MNIST | Pseudo-label · MNIST (n_l=1000) | Calibration (cross-entropy/ECE) | Helped |
| Add debiasing term to Pseudo-label on CIFAR-10 | Pseudo-label · CIFAR-10 (n_l=4000) | Cross entropy and ECE | Helped |
| Add debiasing term to Pseudo-label on DermaMNIST (class-balanced accuracy) | Pseudo-label · DermaMNIST (n_l=1000) | Balanced accuracy: DePseudoLabel 28.84 vs PseudoLabel 22.03 | Helped |
| Add debiasing term to Pseudo-label on MedMNIST | Pseudo-label · MedMNIST five small 2D datasets | Accuracy | Mixed |
| Add debiasing term to Pseudo-label on MedMNIST | Pseudo-label · MedMNIST five small 2D datasets | Cross entropy | Helped |
| Debiasing of consistency methods (VAT, Pi-model, MeanTeacher) | VAT, Pi-model, MeanTeacher · CIFAR-10 and SVHN | Accuracy and cross-entropy | No clear effect |
Closest papers by meaning
- Rethinking Confidence Scores and Thresholds in Pseudolabeling-based SSL (ICML 2025 · similarity 0.90)
- Erasing the Bias: Fine-Tuning Foundation Models for Semi-Supervised Learning (ICML 2024 · similarity 0.90)
- Language-Assisted Debiasing and Smoothing for Foundation Model-Based Semi-Supervised Learning (CVPR 2025 · similarity 0.90)
- CGMatch: A Different Perspective of Semi-supervised Learning (CVPR 2025 · similarity 0.89)
- CDMAD: Class-Distribution-Mismatch-Aware Debiasing for Class-Imbalanced Semi-Supervised Learning (CVPR 2024 · similarity 0.89)
- Contrastive Credibility Propagation for Reliable Semi-supervised Learning (AAAI 2024 · similarity 0.89)
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.