Research map / Label-efficient and robust learning / Semi-supervised learning
Uncertainty-aware Self-training for Few-shot Text 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 |
|---|---|---|---|
| add class-dependent selection with exploration (ablation) | UST (Easy) · Average of 5 datasets | accuracy: 89.78 -> 91.00 | Helped |
| add confident learning (variance-weighted loss) (ablation) | UST (Easy) · Average of 5 datasets | accuracy: 89.96 -> 91.00 | Helped |
| add class-dependent selection with exploration (ablation) | UST (Hard) · Average of 5 datasets | accuracy: 88.27 -> 90.02 | Helped |
| add confident learning (ablation) | UST (Hard) · Average of 5 datasets | accuracy: 89.57 -> 90.02 | Helped |
| uncertainty-based easy sample selection instead of uniform sampling (ablation) | classic self-training · Average of 5 datasets | accuracy: 87.34 -> 91.00 | Helped |
| select easy instead of hard samples (ablation) | UST · Average of 5 datasets | accuracy: 90.02 -> 91.00 | Helped |
| UST vs UDA (no back-translation needed) | BERT-Base few-shot · Average of 5 datasets | accuracy: 89.06 -> 91.00 | Helped |
| UST vs UDA on IMDB | BERT-Base few-shot · IMDB | accuracy: 89.30 -> 89.21 | No clear effect |
| add uncertainty-aware self-training | BERT-Base with 30 labels per class · Average of 5 datasets | accuracy: 80.85 -> 91.00 | Helped |
| remove class-dependent selection (hard variant) on SST (ablation) | UST (Hard) · SST | accuracy: 80.45 -> 88.02 | Helped |
Closest papers by meaning
- Uncertainty-Aware Self-Training for Low-Resource Neural Sequence Labeling (AAAI 2023 · similarity 0.90)
- Tuning Language Models as Training Data Generators for Augmentation-Enhanced Few-Shot Learning (ICML 2023 · similarity 0.88)
- Self Iterative Label Refinement via Robust Unlabeled Learning (NeurIPS 2025 · similarity 0.88)
- Revisiting Self-Training for Neural Sequence Generation (ICLR 2020 · similarity 0.88)
- Neighborhood-Regularized Self-Training for Learning with Few Labels (AAAI 2023 · similarity 0.87)
- Debiased Self-Training for Semi-Supervised Learning (NeurIPS 2022 · 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.