Research map / Label-efficient and robust learning / Recognition and long-tail
Probably Approximately Correct Labels
Emmanuel Candes, Andrew Ilyas, Tijana Zrnic
Read the paperCheck your idea against 137,217 papers
AI-extracted. The measured effects below were read by AI from the paper's full text (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 uncertainty calibration by group and bin (ablation) | PAC labeling · Media bias (GPT-4o) | Budget save (%): 13.68 -> 16.72 | Improved |
| route between GPT-4o and Claude with a PAC router instead of GPT-4o alone | PAC labeling · Media bias | Budget save (%): 13.79 -> 41.61 | Improved |
| route between GPT-4o and Claude with a PAC router instead of Claude alone | PAC labeling · Media bias | Budget save (%): 8.41 -> 41.61 | Improved |
| use cost-sensitive PAC router instead of GPT-4o alone | PAC labeling · Media bias | Save in cost (%): 188.66 -> 482.04 | Improved |
| use cost-sensitive PAC router instead of Claude alone | PAC labeling · Media bias | Save in cost (%): 131.36 -> 482.04 | Improved |
| use asymptotic instead of nonasymptotic confidence intervals (ablation) | PAC labeling · Media bias | Budget save (%): 13.79 -> 16.11 | Mixed |
| use asymptotic confidence intervals (ablation) | PAC labeling · Stance on global warming | Budget save (%): 28.09 -> 32.15 | Mixed |
| use asymptotic confidence intervals (ablation) | PAC labeling · Misinformation | Budget save (%): 18.12 -> 21.41 | Mixed |
| use asymptotic confidence intervals (ablation) | PAC labeling · ImageNet | Budget save (%): 59.64 -> 62.82 | Mixed |
| use asymptotic confidence intervals (ablation) | PAC labeling · ImageNet v2 | Budget save (%): 39.07 -> 39.20 | No clear effect |
| use PAC labeling instead of fixed threshold U>=0.1 | Naive threshold · Stance on global warming | Error (%): 10.13 -> 4.57 | Improved |
| use PAC labeling instead of fixed threshold U>=0.1 | Naive threshold · Misinformation | Error (%): 7.07 -> 3.80 | Improved |
Closest papers by meaning
- The ALCHEmist: Automated Labeling 500x CHEaper than LLM Data Annotators (NeurIPS 2024 · similarity 0.88)
- Label-Efficient Dataset Pruning via Semi-Supervised Pseudo-Labeling (NeurIPS 2026 · similarity 0.86)
- Self Iterative Label Refinement via Robust Unlabeled Learning (NeurIPS 2025 · similarity 0.86)
- Learning from Noisy Labels via Conditional Distributionally Robust Optimization (NeurIPS 2024 · similarity 0.85)
- Pearls from Pebbles: Improved Confidence Functions for Auto-labeling (NeurIPS 2024 · similarity 0.85)
- CGMatch: A Different Perspective of Semi-supervised Learning (CVPR 2025 · similarity 0.85)
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.
