Research map / Label-efficient and robust learning / Semi-supervised learning
Fidelity-Weighted Learning
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 teacher confidence to weight student updates on teacher-generated labels rather than fine-tuning on those labels uniformly (ablation) | FWL student · Robust04 | MAP: 0.2980 -> 0.3124 | Helped |
| Use teacher confidence to weight student updates on teacher-generated labels rather than fine-tuning on those labels uniformly (ablation) | FWL student · ClueWeb | MAP: 0.1386 -> 0.1472 | Helped |
| Pretrain the representation on weakly labeled data instead of using task-independent unsupervised representations (ablation) | FWL · Robust04 | MAP: 0.2211 -> 0.3124 | Helped |
| Pretrain on weak data and fine-tune on strong data rather than training only on strong data (ablation) | Student ranker · Robust04 | MAP: 0.1790 -> 0.2810 | Helped |
| Use fidelity-weighted teacher-generated soft labels rather than fine-tuning on true observations alone | Student · Toy function approximation | RMSE: 0.5451 -> 0.4143 | Helped |
| Use confidence-weighted updates rather than uniform updates over teacher-generated labels (ablation) | FWL · Document ranking and sentiment classification | performance | Helped |
| Use a clustered Gaussian process instead of a single Gaussian process as teacher (ablation) | FWL teacher · Document ranking and sentiment classification | performance | Helped |
| Use a clustered Gaussian process instead of a single GP after PCA (ablation) | FWL teacher · Document ranking and sentiment classification | performance | Helped |
| Increase β when strong observations are extremely scarce, relying more on weak signals (ablation) | FWL · Toy Data * | generalization performance | Helped |
| Decrease β when the weak annotator is extremely poor, putting more focus on strong observations (ablation) | FWL · Toy Data ** | generalization performance | Helped |
| Use confidence-proportional sampling instead of confidence-weighted learning rates (ablation) | FWL_s · Document ranking and sentiment classification | performance over training | Mixed |
Closest papers by meaning
- Trust Functions: Near Lossless Weak-to-Strong Generalization by Learning to Trust the Weak Teacher (ICML 2026 · similarity 0.88)
- Making Deep Neural Networks Robust to Label Noise: A Loss Correction Approach (CVPR 2017 · similarity 0.87)
- From Linear to Nonlinear: Provable Weak-to-Strong Generalization through Feature Learning (NeurIPS 2025 · similarity 0.87)
- Weak-to-Strong Generalization Even in Random Feature Networks, Provably (ICML 2025 · similarity 0.86)
- Training Deep Neural Networks with Virtual Smoothing Classes (AAAI 2025 · similarity 0.86)
- Learning with Structural Labels for Learning with Noisy Labels (CVPR 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.