Research map / Label-efficient and robust learning / Semi-supervised learning
Meta Pseudo Labels
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 |
|---|---|---|---|
| Adapt the teacher using feedback from the student's labeled-data performance, rather than keeping the teacher fixed | Pseudo-label training on ImageNet · ImageNet ILSVRC 2012 val | top-1 accuracy: 88.6 -> 90.2 | Helped |
| Use Meta Pseudo Labels instead of UDA | ResNet-50 semi-supervised learning · ImageNet-10% | top-1 accuracy: 68.07 -> 73.89 | Helped |
| Use Meta Pseudo Labels instead of UDA | ResNet-50 with JFT unlabeled data · ImageNet ILSVRC 2012 val | top-1 accuracy: 79.0 -> 83.2 | Helped |
| Use Meta Pseudo Labels instead of Noisy Student + FixRes | EfficientNet-L2 with JFT unlabeled data · ImageNet ILSVRC 2012 val | top-1 accuracy: 88.5 -> 90.2 | Helped |
| Use Meta Pseudo Labels instead of the strongest listed baseline | WideResNet-28-2 semi-supervised learning · CIFAR-10-4K | accuracy: 95.74 ± 0.05 -> 96.11 ± 0.07 | Helped |
| Use Meta Pseudo Labels instead of the strongest listed baseline | WideResNet-28-2 semi-supervised learning · SVHN-1K | accuracy: 97.72 ± 0.38 -> 98.01 ± 0.07 | Helped |
| Use feedback to adjust the teacher's pseudo labels | Pseudo-label learning on TwoMoon · TwoMoon | success finding the correct classifier | Helped |
| Add supervised and UDA auxiliary objectives to the teacher (ablation) | Meta Pseudo Labels · Image classification experiments | performance | Helped |
| Use soft rather than hard pseudo labels (ablation) | Meta Pseudo Labels · Small-scale experiments | performance | No clear effect |
| Add a moving-average baseline when computing the teacher feedback coefficient (ablation) | Meta Pseudo Labels · Image classification experiments | performance | Helped |
| Replace the gradient dot product with cosine similarity for the feedback coefficient (ablation) | Meta Pseudo Labels · Training stability | numerical stability | Helped |
Closest papers by meaning
- Curriculum Labeling: Revisiting Pseudo-Labeling for Semi-Supervised Learning (AAAI 2021 · similarity 0.91)
- All Labels Are Not Created Equal: Enhancing Semi-Supervision via Label Grouping and Co-Training (CVPR 2021 · similarity 0.90)
- Barely-Supervised Learning: Semi-supervised Learning with Very Few Labeled Images (AAAI 2022 · similarity 0.88)
- Learning to Self-Train for Semi-Supervised Few-Shot Classification (NeurIPS 2019 · similarity 0.88)
- Diffusion Models and Semi-Supervised Learners Benefit Mutually with Few Labels (NeurIPS 2023 · similarity 0.88)
- ProtoCon: Pseudo-Label Refinement via Online Clustering and Prototypical Consistency for Efficient Semi-Supervised Learning (CVPR 2023 · 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.