Research map / Label-efficient and robust learning / Semi-supervised learning
Learning to Self-Train for Semi-Supervised Few-Shot Classification
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 LST with recursive self-training, hard selection, and SWN instead of MTL | MTL / ResNet-12 · miniImageNet, 1-shot | accuracy: 61.2 -> 70.1 | Helped |
| apply LST with recursive self-training, hard selection, and SWN instead of MTL | MTL / ResNet-12 · tieredImageNet, 1-shot | accuracy: 65.6 -> 77.7 | Helped |
| add hard selection of confident pseudo-labels (ablation) | self-training without selection · miniImageNet, 1-shot | accuracy: 59.7 -> 63.0 | Helped |
| add recursive self-training (ablation) | hard selection without meta-training · miniImageNet, 1-shot | accuracy: 63.0 -> 64.6 | Helped |
| add meta-learning of Φss and θ′ (ablation) | hard selection · tieredImageNet, 1-shot | accuracy: 69.8 -> 74.7 | Helped |
| add soft weighting with SWN to hard selection (ablation) | meta hard selection · miniImageNet, 1-shot | accuracy: 64.1 -> 65.0 | Helped |
| add recursive self-training to hard selection and soft weighting (ablation) | LST · miniImageNet, 1-shot | accuracy: 65.0 -> 70.1 | Helped |
| mix all unlabeled subsets for one retraining round instead of recursive training (ablation) | hard selection and soft weighting · miniImageNet, 1-shot | accuracy: 70.1 -> 66.2 | Hurt |
| add recursive self-training (ablation) | hard selection without meta-training · miniImageNet with distractors, 1-shot | accuracy: 61.6 -> 61.2 | Hurt |
| add recursive self-training (ablation) | hard selection without meta-training · tieredImageNet with distractors, 1-shot | accuracy: 68.8 -> 68.3 | Hurt |
| use recursive self-training rather than mixing subsets into one round (ablation) | hard selection and soft weighting · miniImageNet with distractors, 1-shot | accuracy: 64.5 vs 64.1 | Mixed |
| use more retraining steps (ablation) | LST · miniImageNet, 1-shot | accuracy | Mixed |
Closest papers by meaning
- Meta-Learning for Semi-Supervised Few-Shot Classification (ICLR 2018 · similarity 0.93)
- An Embarrassingly Simple Approach to Semi-Supervised Few-Shot Learning (NeurIPS 2022 · similarity 0.92)
- PLATINUM: Semi-Supervised Model Agnostic Meta-Learning using Submodular Mutual Information (ICML 2022 · similarity 0.90)
- Barely-Supervised Learning: Semi-supervised Learning with Very Few Labeled Images (AAAI 2022 · similarity 0.90)
- Big Self-Supervised Models are Strong Semi-Supervised Learners (NeurIPS 2020 · similarity 0.90)
- Pushing the Limit of Fine-Tuning for Few-Shot Learning: Where Feature Reusing Meets Cross-Scale Attention (AAAI 2024 · 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.