Research map / Label-efficient and robust learning / Noisy labels
Neighborhood-Regularized Self-Training for Learning with Few Labels
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 |
|---|---|---|---|
| Replace prediction-only sample selection with neighborhood-regularized selection and prediction aggregation | self-training · text datasets (four datasets) | performance: 1.83% | Helped |
| Replace prediction-only sample selection with neighborhood-regularized selection and prediction aggregation | self-training · graph datasets (four datasets) | performance: 2.51% | Helped |
| Use NeST selection instead of the best baseline selection | self-training · selected pseudo-labeled data | pseudo-label error: 36.8% | Helped |
| Use NeST instead of UST/CEST selection | self-training · one self-training iteration | running time: 57.3% | Helped |
| Add noise-robust loss functions to NeST (ablation) | NeST · AG News and Elec | accuracy: 87.8 ±0.8 -> 88.3 ±0.5; 89.5 ±0.3 -> 89.8 ±0.2; 92.0 ±0.3 -> 92.3 ±0.2; 93.0 ±0.2 -> 93.1 ±0.3 | Helped |
| Add noise-robust loss functions to CEST (ablation) | CEST · AG News and Elec | accuracy: 86.5 -> 87.1; 91.5 -> 92.2 | Helped |
| Increase the neighborhood size k (ablation) | NeST · NYT and Chemprot (30 labels/class) | performance | Mixed |
| Increase the selected unlabeled-data multiplier c (ablation) | NeST · NYT and Chemprot (30 labels/class) | performance | Mixed |
| Use neighborhood-regularized sample selection (ablation) | NeST · text and graph datasets | performance | Helped |
| Aggregate selection scores across training iterations (ablation) | NeST · text and graph datasets | performance stability | Helped |
| Use confidence-based selection rather than neighborhood-based selection | self-training · molecular property prediction | performance | Hurt |
| Use uncertainty-based selection without neighbor information rather than NeST | self-training · graph datasets | performance | Hurt |
Closest papers by meaning
- Theoretical Analysis of Self-Training with Deep Networks on Unlabeled Data (ICLR 2021 · similarity 0.90)
- Debiased Self-Training for Semi-Supervised Learning (NeurIPS 2022 · similarity 0.90)
- Learning With Neighbor Consistency for Noisy Labels (CVPR 2022 · similarity 0.89)
- Partial Label Learning with Self-Guided Retraining (AAAI 2019 · similarity 0.88)
- Leveraging Local Variance for Pseudo-Label Selection in Semi-supervised Learning (AAAI 2024 · similarity 0.87)
- Self-cognitive Denoising in the Presence of Multiple Noisy Label Sources (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.