Research map / Learning theory and algorithms / Sample complexity
Learning from Noisy Labels with No Change to the Training Process
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 |
|---|---|---|---|
| noise-corrected plug-in applied only at prediction time (no training change) | plug-in with known C · MNIST and CIFAR10 with asymmetric noise | test accuracy | No clear effect |
| noise-corrected plug-in with iterative noise estimation vs. baselines | noise estimation heuristic · MNIST CMNIST(0.45-0.65) | test accuracy | Helped |
| removing noise correction (standard logistic regression) | logistic regression · symmetric noise Csym(0.2) on MNIST and CIFAR10 | test accuracy | No clear effect |
| noise correction (plug-in with known C) added to standard logistic regression (ablation) | heavy asymmetric noise CMNIST(0.6) and CCIFAR10(0.65) · MNIST and CIFAR10 | test accuracy: MNIST 49.86 -> 89.23; CIFAR10 54.79 -> 84.01 | Helped |
| sample size increase with growing ||C^-1||_2 (noise matrix nearer singular) (ablation) | plug-in on synthetic 5-class data · synthetic 5-class, 0-1 and ordinal losses | accuracy / Lord loss | Hurt |
| iterative noise estimation heuristic vs. Patrini anchor estimation | plug-in noise estimation · CMNIST(0.6) | test accuracy: 70.68 vs 69.64 (Patrini, plug-in) on CMNIST(0.6) col; | Helped |
Closest papers by meaning
- Safeguarded Dynamic Label Regression for Noisy Supervision (AAAI 2019 · similarity 0.90)
- Beyond Class-Conditional Assumption: A Primary Attempt to Combat Instance-Dependent Label Noise (AAAI 2021 · similarity 0.89)
- Debiased Sample Selection for Learning with Noisy Labels (CVPR 2026 · similarity 0.89)
- Robustness of Accuracy Metric and its Inspirations in Learning with Noisy Labels (AAAI 2021 · similarity 0.89)
- Estimating Noisy Class Posterior with Part-level Labels for Noisy Label Learning (CVPR 2024 · similarity 0.88)
- Statistical Query Hardness of Multiclass Linear Classification with Random Classification Noise (ICML 2025 · 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.