Research map / Label-efficient and robust learning / Noisy labels
Symmetric Cross Entropy for Robust Learning With Noisy Labels
Yisen Wang, Xingjun Ma, Zaiyi Chen, Yuan Luo, Jinfeng Yi, James Bailey
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 reverse cross entropy (SL) to CE | CE, 8-layer CNN · CIFAR-10 (60% symmetric noise) | test accuracy (%): 66.62 -> 80.07 | Improved |
| add reverse cross entropy (SL) to CE | CE, 8-layer CNN · CIFAR-10 (40% symmetric noise) | test accuracy (%): 78.70 -> 85.34 | Improved |
| add reverse cross entropy (SL) to CE | CE, ResNet-44 · CIFAR-100 (60% symmetric noise) | test accuracy (%): 25.39 -> 41.47 | Improved |
| add reverse cross entropy (SL) to CE | CE, 4-layer CNN · MNIST (60% symmetric noise) | test accuracy (%): 46.54 -> 97.40 | Improved |
| add reverse cross entropy (SL) to CE | CE, ResNet-50 · Clothing1M | accuracy (%): 68.80 -> 71.02 | Improved |
| add SL to CE to improve learning on hard classes (ablation) | CE, 8-layer CNN · CIFAR-10 (40% symmetric noise), class 3 clean portion | prediction confidence: 0.5 -> 0.95 | Improved |
| add RCE term to Forward | Forward, 8-layer CNN · CIFAR-10 (40% symmetric noise) | test accuracy (%): 81.37 -> 84.54 | Improved |
| add RCE term to Forward | Forward, 8-layer CNN · CIFAR-10 (60% symmetric noise) | test accuracy (%): 73.59 -> 79.64 | Improved |
| add RCE term to Forward | Forward, 8-layer CNN · CIFAR-10 (40% asymmetric noise) | test accuracy (%): 85.73 -> 86.22 | Improved |
| replace LSR with SL loss | LSR, 8-layer CNN · CIFAR-10 (40% symmetric noise) | test accuracy (%): 78.41 -> 85.20 | Improved |
| replace LSR with SL loss | LSR, 8-layer CNN · CIFAR-10 (40% asymmetric noise) | test accuracy (%): 77.88 -> 80.99 | Improved |
| upscale the CE term (2CE, 5CE) instead of adding RCE (ablation) | CE, 8-layer CNN · CIFAR-10 (60% symmetric noise) | test accuracy | Hurt |
Closest papers by meaning
- Learning from Noisy Labels with Complementary Loss Functions (AAAI 2021 · similarity 0.92)
- Learning With Noisy Labels via Sparse Regularization (ICCV 2021 · similarity 0.90)
- Generalized Cross Entropy Loss for Training Deep Neural Networks with Noisy Labels (NeurIPS 2018 · similarity 0.90)
- $\epsilon$-Softmax: Approximating One-Hot Vectors for Mitigating Label Noise (NeurIPS 2024 · similarity 0.87)
- Training Deep Neural Networks with Virtual Smoothing Classes (AAAI 2025 · similarity 0.87)
- Learning with Structural Labels for Learning with Noisy Labels (CVPR 2024 · similarity 0.86)
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.
