Research map / Label-efficient and robust learning / Noisy labels
Asymmetric Loss Functions for Learning with Noisy Labels
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 asymmetric loss functions instead of conventional baselines | MNIST · symmetric label noise 0.8 | test accuracy (%): 93.78 ± 0.41 -> 96.73 ± 0.20 | Helped |
| Combine asymmetric loss AUL with NCE | CIFAR-10 · symmetric label noise 0.8 | test accuracy (%): 55.74 ± 0.95 -> 59.44 ± 1.14 | Helped |
| Use NCE+AGCE instead of NCE+RCE | CIFAR-100 · symmetric label noise 0.8 | test accuracy (%): 25.41 ± 0.98 -> 24.72 ± 0.60 | Hurt |
| Use AUL instead of the strongest listed baseline | MNIST · asymmetric label noise 0.4 | test accuracy (%): 95.84 ± 0.26 -> 98.62 ± 0.09 | Helped |
| Combine AGCE with NCE instead of using NCE+RCE | CIFAR-10 · asymmetric label noise 0.4 | test accuracy (%): 79.33 ± 0.15 -> 80.00 ± 0.44 | Helped |
| Combine AGCE with NCE instead of using NCE+RCE | CIFAR-100 · asymmetric label noise 0.4 | test accuracy (%): 41.68 ± 0.56 -> 43.76 ± 0.70 | Helped |
| Use AGCE instead of the strongest listed baseline | WebVision · validation, 50 concepts | Top-1 validation accuracy (%): 66.96 -> 69.40 | Helped |
| Increase the asymmetry ratio through larger hyperparameter a or ratio r (ablation) | AGCE, AUL, and AEL · CIFAR-10, symmetric noise 0.8 | robustness to label noise | Mixed |
| Set AGCE q=0.5 with asymmetry ratio below 16/65 (ablation) | AGCE · CIFAR-10, symmetric noise 0.8 | robustness to label noise | Hurt |
| Use AGCE q=1.5 rather than the lower-asymmetry setting (ablation) | AGCE · CIFAR-10, symmetric noise 0.8 | robustness to label noise | Helped |
| Use AUL with p<1 rather than p=2 (ablation) | AUL · CIFAR-10, symmetric noise 0.8 | robustness to label noise | Helped |
Closest papers by meaning
- Normalized Loss Functions for Deep Learning with Noisy Labels (ICML 2020 · similarity 0.94)
- Robust Loss Functions under Label Noise for Deep Neural Networks (AAAI 2017 · similarity 0.94)
- Enhancing Noise-Robust Losses for Large-Scale Noisy Data Learning (AAAI 2025 · similarity 0.91)
- Learning from Noisy Labels with Complementary Loss Functions (AAAI 2021 · similarity 0.90)
- $\epsilon$-Softmax: Approximating One-Hot Vectors for Mitigating Label Noise (NeurIPS 2024 · similarity 0.89)
- Robust Loss Functions for Training Decision Trees with Noisy Labels (AAAI 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.