Research map / Label-efficient and robust learning / Noisy labels
A Gift from Label Smoothing: Robust Training with Adaptive Label Smoothing via Auxiliary Classifier under Label Noise
Jongwoo Ko, Bongsoo Yi, Se-Young Yun
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 ALASCA to Standard training | Standard · CIFAR-10, symmetric 20% noise | accuracy: 87.0 $\pm$ 0.1 -> 92.2 $\pm$ 0.2 | Improved |
| Add ALASCA to Co-teaching | Co-teaching · CIFAR-100, symmetric 80% noise | accuracy: 20.5 $\pm$ 1.3 -> 42.2 $\pm$ 1.2 | Improved |
| Add ALASCA to ELR+ | ELR+ · Clothing1M | accuracy: 74.81 -> 74.92 | Improved |
| Add auxiliary classifiers (ablation) | Uniform label smoothing · CIFAR-10, symmetric 20% noise | accuracy: 87.3 $\pm$ 0.5 -> 91.1 $\pm$ 0.1 | Improved |
| Add adaptive rather than uniform label smoothing (ablation) | Label smoothing with auxiliary classifiers · CIFAR-10, symmetric 20% noise | accuracy: 91.1 $\pm$ 0.1 -> 91.8 $\pm$ 0.1 | Improved |
| Add EMA confidence to adaptive label smoothing without auxiliary classifiers (ablation) | ALS · CIFAR-10, asymmetric 40% noise | accuracy: 85.1 $\pm$ 0.2 -> 86.9 $\pm$ 1.1 | Mixed |
| Add EMA confidence to ALS + AC (ablation) | ALS + AC · CIFAR-10, symmetric 20% noise | accuracy: 91.8 $\pm$ 0.1 -> 92.2 $\pm$ 0.2 | Improved |
| Ensemble predictions from the main and auxiliary classifiers at inference (ablation) | ALASCA · CIFAR-10, symmetric 20% noise | accuracy: 92.2 $\pm$ 0.2 -> 92.8 $\pm$ 0.1 | Improved |
| Use bottleneck rather than MLP auxiliary classifiers at all blocks (ablation) | ALASCA · CIFAR-10, instance-dependent 40% noise | accuracy: 80.8 $\pm$ 0.3 -> 77.9 $\pm$ 0.4 | Hurt |
Closest papers by meaning
- From Label Smoothing to Label Relaxation (AAAI 2021 · similarity 0.88)
- MaxSup: Overcoming Representation Collapse in Label Smoothing (NeurIPS 2025 · similarity 0.87)
- $\epsilon$-Softmax: Approximating One-Hot Vectors for Mitigating Label Noise (NeurIPS 2024 · similarity 0.86)
- On the Role of Label Noise in the Feature Learning Process (ICML 2025 · similarity 0.86)
- For Better or For Worse? Learning Minimum Variance Features With Label Augmentation (ICLR 2025 · similarity 0.86)
- NoiseGPT: Label Noise Detection and Rectification through Probability Curvature (NeurIPS 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.
