Research map / Label-efficient and robust learning / Noisy labels
Suppressing Mislabeled Data via Grouping and Self-Attention
Xiaojiang Peng, Kai Wang, Zhaoyang Zeng, Qing Li, Jianfei Yang, Yu Qiao
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 AFM on top of ResNet50 baseline | ResNet50 · Food101N (test on Food-101) | accuracy (%): 84.51 -> 87.23 | Improved |
| add AFM on top of ResNet50 baseline | ResNet50 · Clothing1M (1M noisy only) | accuracy (%): 68.94 -> 74.12 | Improved |
| use sum interaction instead of concatenation in Group-to-Attend (ablation) | AFM · Food101N (test on Food-101) | accuracy (%): 86.95 -> 87.23 | Improved |
| use sum instead of element-wise multiplication interaction (ablation) | AFM · Food101N (test on Food-101) | accuracy (%): 86.64 -> 87.23 | Improved |
| add separate FC projections FC_a and FC_b before sum interaction (ablation) | AFM · Food101N (test on Food-101) | accuracy (%): 86.12 -> 87.23 | Improved |
| add separate FC projections before concatenation (ablation) | AFM · Food101N (test on Food-101) | accuracy (%): 86.51 -> 86.95 | Improved |
| add the mixup loss term (lambda 0 -> 0.75) (ablation) | AFM · Food101N (test on Food-101) | accuracy (%): 84.51 -> 87.23 | Improved |
| use only interpolations (lambda 1.00) instead of lambda 0.75 mixing in original samples loss (ablation) | AFM · Food101N (test on Food-101) | accuracy (%): 87.23 -> 86.47 | Hurt |
| increase group size from 2 to 6 (ablation) | AFM · Food101N (test on Food-101) | accuracy (%): 87.23 -> 85.46 | Hurt |
| use learned attention weights (AFM) instead of standard mixup (ablation) | AFM · Food101N (test on Food-101) | accuracy (%): 85.63 -> 87.23 | Improved |
| use learned attention weights (AFM) instead of Manifold mixup (ablation) | AFM · Food101N (test on Food-101) | accuracy (%): 86.12 -> 87.23 | Improved |
| share classifiers f_c1 and f_c2 (ablation) | AFM · Food101N (test on Food-101) | accuracy (%): 86.97 -> 87.23 | Improved |
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.
- MetaCleaner — MetaCleaner: Learning to Hallucinate Clean Representations for Noisy-Labeled Visual Recognition (CVPR 2019) · better on 4 of 4 shared results
- CurriculumNet — CurriculumNet: Weakly Supervised Learning from Large-Scale Web Images (ECCV 2018) · 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
- C-Mixup: Improving Generalization in Regression (NeurIPS 2022 · similarity 0.87)
- Catch-Up Mix: Catch-Up Class for Struggling Filters in CNN (AAAI 2024 · similarity 0.86)
- IOFM: Using the Interpolation Technique on the Over-Fitted Models to Identify Clean-Annotated Samples (AAAI 2024 · similarity 0.86)
- Let Samples Speak: Mitigating Spurious Correlation by Exploiting the Clusterness of Samples (CVPR 2025 · similarity 0.85)
- A Simple Remedy for Dataset Bias via Self-Influence: A Mislabeled Sample Perspective (NeurIPS 2024 · similarity 0.85)
- Adversarial AutoMixup (ICLR 2024 · similarity 0.84)
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.
