Research map / Label-efficient and robust learning / Semi-supervised learning
Not All Parameters Should Be Treated Equally: Deep Safe Semi-supervised Learning under Class Distribution Mismatch
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 |
|---|---|---|---|
| Apply SPL to CL | CL · CIFAR-10, mismatch ratios 0.3 and 0.6 | accuracy (%): 83.2±0.4 -> 87.8±0.3; 82.1±0.4 -> 84.1±0.5 | Helped |
| Apply SPL to CL | CL · CIFAR-100, mismatch ratios 0.3 and 0.6 | accuracy (%): 63.6±0.4 -> 65.9±0.3; 61.5±0.5 -> 65.5±0.4 | Helped |
| Apply SPL to CL | CL · TinyImageNet, mismatch ratios 0.3 and 0.6 | accuracy (%): 37.3±0.7 -> 38.6±0.5; 36.7±0.8 -> 37.7±0.5 | Helped |
| Use soft mask instead of hard mask (ablation) | SPL with CL on CIFAR-100 · CIFAR-100, mismatch ratios 0.3–1.0 | accuracy (%): 65.9±0.3 -> 66.9±0.4; 65.5±0.4 -> 66.7±0.2; 60.9±0.2 -> 63.6±0.3; 60.7±0.4 -> 63.5±0.5 | Helped |
| Remove pre-training (ablation) | SPL with VAT on CIFAR-10 · CIFAR-10, mismatch ratio=0.3, m=400×6 | accuracy (%): 80.1±0.2 -> 33.8±1.1 | Hurt |
| Add safe-parameter identification (ablation) | SPL with VAT on CIFAR-10 · CIFAR-10, mismatch ratio=0.3, m=400×6 | accuracy (%): 77.0±0.5 -> 80.1±0.2 | Helped |
| Add bi-level optimization (ablation) | SPL with VAT on CIFAR-10 · CIFAR-10, mismatch ratio=0.3, m=400×6 | accuracy (%): 77.2±0.6 -> 80.1±0.2 | Helped |
| Add inner-level optimization (ablation) | SPL with VAT on CIFAR-10 · CIFAR-10, mismatch ratio=0.3, m=400×6 | accuracy (%): 76.3±0.3 -> 80.1±0.2 | Helped |
| Add outer-level optimization (ablation) | SPL with VAT on CIFAR-10 · CIFAR-10, mismatch ratio=0.3, m=400×6 | accuracy (%): 79.8±0.2 -> 80.1±0.2 | Helped |
| Fine-tune CL directly using the validation subset V1 (ablation) | CL on CIFAR-100 · CIFAR-100, mismatch ratios 0.3–1.0 | accuracy (%): 63.6±0.4 -> 62.1±0.5; 63.2±0.2 -> 62.0±0.2; 62.4±0.3 -> 61.1±0.2; 61.5±0.5 -> 61.0±0.5; 61.5±0.4 -> 59.2±0.3; 59.3±0.2 - | Hurt |
Closest papers by meaning
- Safe Deep Semi-Supervised Learning for Unseen-Class Unlabeled Data (ICML 2020 · similarity 0.94)
- Safe-Student for Safe Deep Semi-Supervised Learning With Unseen-Class Unlabeled Data (CVPR 2022 · similarity 0.91)
- Re-Evaluating the Impact of Unseen-Class Unlabeled Data on Semi-Supervised Learning Model (ICLR 2025 · similarity 0.87)
- Erasing the Bias: Fine-Tuning Foundation Models for Semi-Supervised Learning (ICML 2024 · similarity 0.85)
- Contrastive Credibility Propagation for Reliable Semi-supervised Learning (AAAI 2024 · similarity 0.84)
- Rethinking Confidence Scores and Thresholds in Pseudolabeling-based SSL (ICML 2025 · similarity 0.84)
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.