Research map / Label-efficient and robust learning / Semi-supervised learning
S4L: Self-Supervised Semi-Supervised Learning
Xiaohua Zhai, Avital Oliver, Alexander Kolesnikov, Lucas Beyer
Read the paperCheck your idea against 137,217 papers
AI-extracted. The measured effects below were read by AI from the paper's full text (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 S4L rotation self-supervised loss | supervised ResNet50v2 · ILSVRC-2012 10% labels | top-5 accuracy (%): 80.43 -> 83.82 | Improved |
| add S4L rotation self-supervised loss | supervised ResNet50v2 · ILSVRC-2012 1% labels | top-5 accuracy (%): 48.43 -> 53.37 | Improved |
| add S4L exemplar self-supervised loss | supervised ResNet50v2 · ILSVRC-2012 1% labels | top-5 accuracy (%): 48.43 -> 47.02 | Hurt |
| add S4L exemplar self-supervised loss | supervised ResNet50v2 · ILSVRC-2012 10% labels | top-5 accuracy (%): 80.43 -> 83.72 | Improved |
| add VAT loss | supervised ResNet50v2 · ILSVRC-2012 1% labels | top-5 accuracy (%): 48.43 -> 44.05 | Hurt |
| add entropy minimization to VAT | VAT · ILSVRC-2012 10% labels | top-5 accuracy (%): 82.78 -> 83.39 | Improved |
| add pseudo-label retraining step | MOAM step 1 · ILSVRC-2012 10% labels | top-5 accuracy (%): 88.80 -> 89.96 | Improved |
| add final fine-tuning on the 10% labels | MOAM + pseudo label · ILSVRC-2012 10% labels | top-5 accuracy (%): 89.96 -> 91.23 | Improved |
| apply supervised loss on rotated copies of labeled images (ablation) | S4L-Rotation · ILSVRC-2012 | top-5 accuracy (%): almost 1% | Improved |
| apply self-supervised loss also on labeled images (ablation) | S4L-Rotation · ILSVRC-2012 | top-5 accuracy (%) | No clear effect |
| average predictions over 4 rotations at inference (ablation) | S4L-Rotation · ILSVRC-2012 | top-5 accuracy (%): 1-2% more accurate | Improved |
| fine-tune 4x-wider supervised vs add MOAM losses with all labels | ResNet50v2 4x wider · ILSVRC-2012 100% labels | top-5 accuracy (%): 94.10 -> 94.97 | Improved |
Closest papers by meaning
- Semi-Supervised Learning Made Simple With Self-Supervised Clustering (CVPR 2023 · similarity 0.90)
- Semi-Supervised and Unsupervised Deep Visual Learning: A Survey (TPAMI 2022 · similarity 0.89)
- Towards Realistic Semi-Supervised Learning (ECCV 2022 · similarity 0.88)
- Revisiting Semi-Supervised Learning in the Era of Foundation Models (NeurIPS 2025 · similarity 0.87)
- Systematic Comparison of Semi-supervised and Self-supervised Learning for Medical Image Classification (CVPR 2024 · similarity 0.86)
- InterLUDE: Interactions between Labeled and Unlabeled Data to Enhance Semi-Supervised Learning (ICML 2024 · similarity 0.86)
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.
