Research map / Label-efficient and robust learning / Noisy labels
Dynamic Loss for Robust Learning
Shenwang Jiang, Jianan Li, Jizhou Zhang, Ying Wang, Tingfa Xu
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 label corrector (ablation) | Dynamic Loss (margin generator only) · CIFAR-10-N-LT (avg last) | accuracy (%): 71.51 -> 79.73 | Improved |
| add margin generator (ablation) | label corrector only · CIFAR-10-N-LT (avg last) | accuracy (%): 71.08 -> 79.73 | Improved |
| add Balanced-Softmax to label corrector only (ablation) | label corrector · CIFAR-10-N-LT (avg last) | accuracy (%): 71.08 -> 72.64 | Improved |
| use hierarchical sampling instead of naive random sampling (ablation) | Dynamic Loss · CIFAR-10-N-LT (avg last) | accuracy (%): 78.53 -> 79.73 | Improved |
| use class-specific label corrector instead of class-agnostic (ablation) | Dynamic Loss · CIFAR-10-N (40% asymmetric) | accuracy (%): 90.56 -> 94.51 | Improved |
| use explicit given/predicted labels instead of implicit (ablation) | Dynamic Loss · CIFAR-10-N-LT (avg last) | accuracy (%): 57.21 -> 79.73 | Improved |
| use group-wise instead of sample-wise optimization for both modules (ablation) | Dynamic Loss · CIFAR-10-N-LT (avg last) | accuracy (%): 44.63 -> 79.73 | Improved |
| use group-wise label corrector in addition to group-wise margin generator (ablation) | Dynamic Loss · CIFAR-10-N-LT (avg last) | accuracy (%): 40.76 -> 79.73 | Improved |
| add MC-Dropout uncertainty to loss rank input (ablation) | meta net · CIFAR-10-N-LT (avg last) | accuracy (%): 79.73 -> 80.62 | Improved |
| use loss rank instead of MC-Dropout uncertainty as input (ablation) | meta net · CIFAR-10-N-LT (avg last) | accuracy (%): 72.91 -> 79.73 | Improved |
| use 3-layer MLP meta net instead of learnable vectors (ablation) | meta net · CIFAR-10-N-LT (avg last) | accuracy (%): 79.50 -> 79.73 | Improved |
| add Balanced-Softmax as base margin to Dynamic Loss | Dynamic Loss · CIFAR-10-LT (avg) | accuracy (%): 87.24 -> 87.34 | No clear effect |
Earlier methods this paper compares with
Named baselines from earlier accepted papers in this paper's results tables: better than 2 of 2 on most shared results.
- DivideMix — DivideMix: Learning with Noisy Labels as Semi-supervised Learning (ICLR 2020) · better on 4 of 4 shared results
- CMW-Net — CMW-Net: Learning a Class-Aware Sample Weighting Mapping for Robust Deep Learning (TPAMI 2023) · better on 1 of 1 shared result
Matched by method name in results tables read by AI. Authors' own tables tend to favour the newer method, and a comparison can be missed when a method is renamed.
Closest papers by meaning
- Robust Classification via Regression for Learning with Noisy Labels (ICLR 2024 · similarity 0.90)
- Label Distributionally Robust Losses for Multi-class Classification: Consistency, Robustness and Adaptivity (ICML 2023 · similarity 0.89)
- Student Loss: Towards the Probability Assumption in Inaccurate Supervision (TPAMI 2024 · similarity 0.89)
- Regroup Median Loss for Combating Label Noise (AAAI 2024 · similarity 0.87)
- Learning from Noisy Labels via Conditional Distributionally Robust Optimization (NeurIPS 2024 · similarity 0.87)
- Forming Auxiliary High-confident Instance-level Loss to Promote Learning from Label Proportions (CVPR 2025 · 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.
