Research map / Label-efficient and robust learning / Noisy labels
Delving into Noisy Label Detection with Clean Data
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 BH procedure with clean-data empirical p-values (BHN) versus prior noisy-only detectors | noisy label detection · CIFAR-10 Asym. 0.3 | FDR: 28.48% lower than best baseline | Helped |
| use BHN (clean-data BH procedure) versus baselines for detection F1 | noisy label detection · CIFAR-10 (ResNet34) | F1: outperforms best baseline by 18.99% | Helped |
| use BHN detected-clean examples to train final classifier versus SimiFeat-V detection | final classifier accuracy · CIFAR-100 Symm. 0.6 | test accuracy: surpasses best baseline by 12.24% | Helped |
| BHN detection and cleaning applied to Clothing1M real noisy labels | final classifier accuracy · Clothing1M | test accuracy: outperforms runner-up by 2.26% | Helped |
| replace negative cross-entropy score with cosine-similarity score (BHN-cosin) (ablation) | BHN score function · CIFAR-10 and CIFAR-100 (ResNet18) | F1: nearly identical (e.g., 95.89 vs 96.20 on CIFAR-10 Symm. 0.6) | No clear effect |
| increase test accuracy of the network used in the score function (ablation) | BHN score function · CIFAR-100 Symm. 0.6 | F1 | Helped |
| lower noise rate (fewer corrupted examples) makes detection harder (ablation) | noisy label detection baselines · CIFAR-10 noise rate 0.1 to 0.6 | F1: SimiFeat-V F1 from 51.37% (η=0.1) to 88.34% (η=0.6) | Hurt |
| BHN versus SimiFeat-V at low noise rate 0.1 (ablation) | noisy label detection · CIFAR-10 noise rate 0.1 | F1: outperforms SimiFeat-V by 33.16% | Helped |
Closest papers by meaning
- FINE Samples for Learning with Noisy Labels (NeurIPS 2021 · similarity 0.90)
- Toward Robustness against Label Noise in Training Deep Discriminative Neural Networks (NeurIPS 2017 · similarity 0.90)
- NoiseGPT: Label Noise Detection and Rectification through Probability Curvature (NeurIPS 2024 · similarity 0.89)
- On the Role of Label Noise in the Feature Learning Process (ICML 2025 · similarity 0.88)
- Handling Label Noise via Instance-Level Difficulty Modeling and Dynamic Optimization (NeurIPS 2025 · similarity 0.88)
- Learning Discriminative Dynamics with Label Corruption for Noisy Label Detection (CVPR 2024 · 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.