Research map / Label-efficient and robust learning / Noisy labels
Generalized Cross Entropy Loss for Training Deep Neural Networks 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 truncated Lq instead of Forward T | CIFAR-10 with uniform noise η=0.4 · CIFAR-10, uniform noise η=0.4 | test accuracy: 85.07 \pm 0.29 -> 87.62 \pm 0.26 | Helped |
| Use truncated Lq instead of Forward T | CIFAR-100 with uniform noise η=0.4 · CIFAR-100, uniform noise η=0.4 | test accuracy: 54.65 \pm 0.88 -> 62.64 \pm 0.33 | Helped |
| Use truncated Lq instead of Forward T | CIFAR-10 with class-dependent noise η=0.4 · CIFAR-10, class-dependent noise η=0.4 | test accuracy: 88.12 \pm 0.32 -> 82.28 \pm 0.67 | Hurt |
| Use truncated Lq instead of Forward T | CIFAR-100 with class-dependent noise η=0.4 · CIFAR-100, class-dependent noise η=0.4 | test accuracy: 70.82 \pm 0.45 -> 47.66 \pm 0.69 | Hurt |
| Use truncated Lq instead of Forward T | FASHION-MNIST with uniform noise η=0.4 · FASHION-MNIST, uniform noise η=0.4 | test accuracy: 92.69 \pm 0.20 -> 92.60 \pm 0.17 | No clear effect |
| Use truncated Lq instead of Forward T | FASHION-MNIST with class-dependent noise η=0.4 · FASHION-MNIST, class-dependent noise η=0.4 | test accuracy: 93.65 \pm 0.11 -> 91.62 \pm 0.34 | Hurt |
| Add truncation to Lq (ablation) | Lq on CIFAR-10 with uniform noise η=0.4 · CIFAR-10, uniform noise η=0.4 | test accuracy: 87.13 \pm 0.22 -> 87.62 \pm 0.26 | No clear effect |
| Add truncation to Lq (ablation) | Lq on CIFAR-100 with uniform noise η=0.4 · CIFAR-100, uniform noise η=0.4 | test accuracy: 61.77 \pm 0.24 -> 62.64 \pm 0.33 | Helped |
| Add truncation to Lq (ablation) | Lq on CIFAR-10 with class-dependent noise η=0.4 · CIFAR-10, class-dependent noise η=0.4 | test accuracy: 76.74\pm 0.61 -> 82.28 \pm 0.67 | Helped |
| Increase q from CCE-like values toward 1 (ablation) | Lq training on noisy CIFAR-10 · CIFAR-10 with uniform label noise | test accuracy | Helped |
| Increase q on clean data (ablation) | Lq training on CIFAR-10 · CIFAR-10 clean data | test accuracy | Hurt |
| Set q=0.8 instead of the standard q=0.7 (ablation) | Lq training with uniform noise η=0.2 · CIFAR-10 with uniform noise η=0.2 | overfitting | Helped |
Closest papers by meaning
- Normalized Loss Functions for Deep Learning with Noisy Labels (ICML 2020 · similarity 0.91)
- Learning from Noisy Labels with Complementary Loss Functions (AAAI 2021 · similarity 0.90)
- Generalized Jensen-Shannon Divergence Loss for Learning with Noisy Labels (NeurIPS 2021 · similarity 0.90)
- $\epsilon$-Softmax: Approximating One-Hot Vectors for Mitigating Label Noise (NeurIPS 2024 · similarity 0.87)
- Enhancing Noise-Robust Losses for Large-Scale Noisy Data Learning (AAAI 2025 · similarity 0.87)
- Handling Label Noise via Instance-Level Difficulty Modeling and Dynamic Optimization (NeurIPS 2025 · similarity 0.85)
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.