Research map / Label-efficient and robust learning / Noisy labels
ProSelfLC: Progressive Self Label Correction for Training Robust Deep Neural Networks
Xinshao Wang, Yang Hua, Elyor Kodirov, David A. Clifton, Neil M. Robertson
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 |
|---|---|---|---|
| Use progressive, entropy-dependent self label correction instead of CCE | ResNet-44 · CIFAR-100 asymmetric noise 40% | accuracy (%): 59.5 -> 67.9 | Improved |
| Use progressive, entropy-dependent self label correction instead of CCE | ResNet-44 · CIFAR-100 symmetric noise 60% | accuracy (%): 37.9 -> 47.7 | Improved |
| Use progressive, entropy-dependent self label correction instead of CCE | ResNet-50 · Clothing 1M | accuracy (%): 72.2 -> 73.4 | Improved |
| Use progressive, entropy-dependent self label correction instead of CCE | ResNet-50 · ImageNet 2012 | accuracy (%): 75.5 -> 76.0 | Improved |
| Increase ProSelfLC growth parameter B from 8 to 10 (ablation) | ProSelfLC · CIFAR-100 clean test | accuracy (%): 70.1 -> 70.3 | Improved |
| Increase ProSelfLC growth parameter B from 10 to 12 (ablation) | ProSelfLC · CIFAR-100 clean test | accuracy (%): 70.3 -> 69.8 | Hurt |
| Set B=16 rather than B=8 (ablation) | ProSelfLC · CIFAR-100 symmetric noise 60% | accuracy (%): 47.7 -> 32.3 | Hurt |
| Raise fixed self-correction weight ε from 0.125 to 0.50 (ablation) | Boot-soft · CIFAR-100 clean test | accuracy (%): 68.9 -> 69.1 | Improved |
| Raise label-smoothing weight ε from 0.125 to 0.50 (ablation) | LS · CIFAR-100 symmetric noise 60% | accuracy (%): 39.1 -> 46.5 | Improved |
| Raise confidence-penalty weight ε from 0.125 to 0.50 (ablation) | CP · CIFAR-100 symmetric noise 60% | accuracy (%): 39.6 -> 44.1 | Improved |
| Use ProSelfLC rather than CCE | CIFAR-100 standard training · CIFAR-100 clean test | accuracy (%): 69.0 -> 70.3 | Improved |
Closest papers by meaning
- Training Noise-Robust Deep Neural Networks via Meta-Learning (CVPR 2020 · similarity 0.86)
- Learning to Purify Noisy Labels via Meta Soft Label Corrector (AAAI 2021 · similarity 0.86)
- Dirichlet-Based Prediction Calibration for Learning with Noisy Labels (AAAI 2024 · similarity 0.86)
- Continuous Review and Timely Correction: Enhancing the Resistance to Noisy Labels via Self-Not-True and Class-Wise Distillation (TPAMI 2025 · similarity 0.86)
- Learning From Noisy Labels With Decoupled Meta Label Purifier (CVPR 2023 · similarity 0.86)
- Two Wrongs Don’t Make a Right: Combating Confirmation Bias in Learning with Label Noise (AAAI 2023 · 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.
