Research map / Label-efficient and robust learning / Semi-supervised learning
A Realistic Evaluation of Semi-Supervised Learning for Fine-Grained Classification
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 MoCo self-supervised pretraining before self-training | ResNet-50 trained from scratch · Semi-Aves, U_in | Top1 accuracy | Helped |
| Add out-of-class unlabeled data to MoCo pretraining | MoCo trained from scratch · Semi-Aves and Semi-Fungi | Top1 accuracy: around 11% | Helped |
| Add out-of-class unlabeled data | FixMatch with expert initialization · Semi-Aves and Semi-Fungi | Top1 accuracy: around 6% | Hurt |
| Add out-of-class unlabeled data | Self-Training with expert initialization · Semi-Aves and Semi-Fungi | Top1 accuracy: less than 1% | No clear effect |
| Use expert initialization instead of training from scratch | Semi-supervised methods · Semi-Aves and Semi-Fungi | Top1 accuracy | Helped |
| Add in-class unlabeled data with FixMatch | ImageNet- or iNat-initialized ResNet-50 · Semi-Aves and Semi-Fungi | Top1 accuracy | Helped |
| Add out-of-class unlabeled data | Pseudo-Label and FixMatch trained from scratch · Semi-Aves and Semi-Fungi | Top1 accuracy: 4-9% | Hurt |
| Add out-of-class unlabeled data | Curriculum Pseudo-Label and Self-Training trained from scratch · Semi-Aves and Semi-Fungi | Top1 accuracy: less than 1% | No clear effect |
| Use a large negative-sample count (65,536) instead of the smaller setting (ablation) | MoCo · Semi-Aves and Semi-Fungi | Top1 accuracy | Hurt |
| Freeze the pretrained feature encoder instead of fine-tuning the entire network (ablation) | MoCo · Semi-Aves and Semi-Fungi | Top1 accuracy | Hurt |
| Increase the pseudo-label confidence threshold (ablation) | Pseudo-Label with expert initialization and out-of-class data · Semi-Aves | Top1 accuracy | Helped |
Closest papers by meaning
- Realistic Evaluation of Deep Semi-Supervised Learning Algorithms (NeurIPS 2018 · similarity 0.91)
- Erasing the Bias: Fine-Tuning Foundation Models for Semi-Supervised Learning (ICML 2024 · similarity 0.89)
- Don’t fear the unlabelled: safe semi-supervised learning via debiasing (ICLR 2023 · similarity 0.89)
- Class-Imbalanced Semi-Supervised Learning with Adaptive Thresholding (ICML 2022 · similarity 0.89)
- Semi-Supervised Learning under Class Distribution Mismatch (AAAI 2020 · similarity 0.88)
- Re-Evaluating the Impact of Unseen-Class Unlabeled Data on Semi-Supervised Learning Model (ICLR 2025 · 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.