Research map / Label-efficient and robust learning / Recognition and long-tail
Uncertainty-Aware Self-Training for Low-Resource Neural Sequence Labeling
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 |
|---|---|---|---|
| Use SeqUST instead of MetaST | 10-shot self-training for sequence labeling · SNIPS, Multiwoz, MIT Movie, MIT Restaurant, CoNLL-03, OntoNotes | F1 (%): 87.33 vs 86.74; 77.98 vs 77.34; 77.73 vs 77.52; 64.19 vs 63.02; 79.10 vs 76.88; 80.33 vs 77.69 | Helped |
| Add token-level reliable-token selection (ablation) | SeqUST · SNIPS, MIT Movie, CoNLL-03 (10-shot) | F1 (%): 82.37 -> 87.33; 70.12 -> 77.73; 73.40 -> 79.10 | Helped |
| Add confidence to token selection (ablation) | SeqUST · SNIPS, MIT Movie, CoNLL-03 (10-shot) | F1 (%): 86.98 -> 87.33; 76.99 -> 77.73; 77.60 -> 79.10 | Helped |
| Add certainty to token selection (ablation) | SeqUST · SNIPS, MIT Movie, CoNLL-03 (10-shot) | F1 (%): 83.33 -> 87.33; 72.29 -> 77.73; 73.16 -> 79.10 | Helped |
| Add PHCE noise-robust loss (ablation) | SeqUST masked sequence labeling · SNIPS, MIT Movie, CoNLL-03 (10-shot) | F1 (%): 87.01 -> 87.33; 77.32 -> 77.73; 78.31 -> 79.10 | Helped |
| Add Gaussian-based consistency regularization (ablation) | SeqUST · SNIPS, MIT Movie, CoNLL-03 (10-shot) | F1 (%): 87.11 -> 87.33; 77.25 -> 77.73; 77.93 -> 79.10 | Helped |
| Combine confidence and certainty instead of using no selection (ablation) | Token selection · SNIPS, MIT Movie, CoNLL-03 unlabeled data | error rate (%): 45.75 -> 17.76; 51.77 -> 23.10; 49.88 -> 24.63 | Helped |
| Combine confidence and certainty instead of confidence alone (ablation) | Token selection · SNIPS, MIT Movie, CoNLL-03 unlabeled data | error rate (%): 43.18 -> 17.76; 48.06 -> 23.10; 45.22 -> 24.63 | Helped |
| Combine confidence and certainty instead of certainty alone (ablation) | Token selection · SNIPS, MIT Movie, CoNLL-03 unlabeled data | error rate (%): 20.42 -> 17.76; 27.60 -> 23.10; 29.10 -> 24.63 | Helped |
Closest papers by meaning
- Learning from Noisy Labels via Conditional Distributionally Robust Optimization (NeurIPS 2024 · similarity 0.86)
- Uncertainty-Aware Self-Training for CTC-Based Automatic Speech Recognition (AAAI 2025 · similarity 0.85)
- Adaptive Integration of Partial Label Learning and Negative Learning for Enhanced Noisy Label Learning (AAAI 2024 · similarity 0.84)
- Self Iterative Label Refinement via Robust Unlabeled Learning (NeurIPS 2025 · similarity 0.84)
- Towards Understanding Self-Pretraining for Sequence Classification (NeurIPS 2026 · similarity 0.84)
- Discrepancy and Uncertainty Aware Denoising Knowledge Distillation for Zero-Shot Cross-Lingual Named Entity Recognition (AAAI 2024 · similarity 0.84)
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.