Research map / Label-efficient and robust learning / Noisy labels
Cost-Sensitive Learning with Noisy Labels
Nagarajan Natarajan, Inderjit S. Dhillon, Pradeep Ravikumar, Ambuj Tewari
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 |
|---|---|---|---|
| increase uniform label noise from 0 to 0.4 (ablation) | C-SVM · banana | accuracy: 90.8 -> 88.5 | Hurt |
| use unbiased logistic loss estimator instead of plain logistic regression | logistic regression · German (linear, rho=0.4) | accuracy: 55.87 -> 67.19 | Improved |
| use unbiased logistic loss estimator instead of plain logistic regression | logistic regression · Breast cancer (linear, rho=0.4) | accuracy: 59.51 -> 67.79 | Improved |
| use unbiased logistic loss estimator instead of plain logistic regression | logistic regression · Heart (linear, rho=0.4) | accuracy: 73.58 -> 57.04 | Hurt |
| use unbiased logistic loss estimator instead of plain logistic regression | logistic regression · Image (linear, rho=0.4) | accuracy: 76.25 -> 63.47 | Hurt |
| tune class weight alpha (C-SVM) instead of equal weights (SVM) (ablation) | SVM · German (linear, rho=0.4) | accuracy: 62.20 -> 56.60 | Hurt |
| tune class weight alpha (C-SVM) instead of equal weights (SVM) (ablation) | SVM · Heart (linear, rho=0.4) | accuracy: 78.27 -> 72.35 | Hurt |
| tune class weight alpha (C-SVM) instead of equal weights (SVM) (ablation) | SVM · UCI sets with equal noise rates | accuracy | No clear effect |
| use C-SVM instead of NHERD | NHERD · Image (linear, rho=0.4) | accuracy: 69.61 -> 73.73 | Improved |
| use C-SVM instead of NHERD | NHERD · Thyroid (linear, rho=0.4) | accuracy: 85.95 -> 76.63 | Hurt |
| optimize AM measure instead of accuracy under class imbalance | accuracy-optimized classifiers · Breast cancer, Diabetes, Thyroid, German | accuracy vs AM | Mixed |
| misspecify noise rates by epsilon in 0.1 to 0.4 (ablation) | batch and online methods · 6 UCI data sets | accuracy ratio | Hurt |
Closest papers by meaning
- Learning from Noisy Labels with No Change to the Training Process (ICML 2021 · similarity 0.91)
- Classification with Noisy Labels by Importance Reweighting (TPAMI 2016 · similarity 0.90)
- Balancing Positive and Negative Classification Error Rates in Positive-Unlabeled Learning (NeurIPS 2025 · similarity 0.88)
- Practical estimation of the optimal classification error with soft labels and calibration (ICLR 2026 · similarity 0.88)
- Class-Weighted Classification: Trade-offs and Robust Approaches (ICML 2020 · similarity 0.87)
- Don’t Label Twice: Quantity Beats Quality when Comparing Binary Classifiers on a Budget (ICML 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.
