Research map / Label-efficient and robust learning / Noisy labels
Combinatorial Inference against Label Noise
Paul Hongsuck Seo, Geeho Kim, Bohyung Han
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 |
|---|---|---|---|
| replace standard flat classifier with combinatorial classification over meta-class sets | ResNet-50 · CUB-200 open-set high uniform noise | accuracy (%): 70.04 -> 79.19 | Improved |
| replace standard classifier with CombCls | ResNet-50 · CUB-200 closed-set high uniform noise | accuracy (%): 42.35 -> 51.90 | Improved |
| replace standard classifier with CombCls on clean data | ResNet-50 · CUB-200 clean | accuracy (%): 80.57 -> 82.80 | Improved |
| combine CombCls with Co-teaching | CombCls · CUB-200 closed-set high uniform noise | accuracy (%): 51.90 -> 62.77 | Improved |
| use q-loss (q=0.5) instead of CombCls under uniform closed-set noise | CombCls · CUB-200 closed-set high uniform noise | accuracy (%): 60.89 vs 51.90 | Hurt |
| k-means clustering instead of random meta-class sets (ablation) | CombCls · CUB-200 closed-set high uniform noise | accuracy (%): 46.98 -> 51.90 | Improved |
| add RL-based meta-class set search (ablation) | CombCls · CUB-200 open-set high uniform noise | accuracy (%): 79.19 -> 79.98 | Improved |
| add RL-based meta-class set search (ablation) | CombCls · WebVision | accuracy (%): 83.14 -> 83.26 | Improved |
| generate meta-class sets from a high-noise dataset rather than clean data (ablation) | CombCls · CUB-200 closed-set | accuracy (%): 45.40 -> 51.90 | Improved |
| generate meta-class sets from a high-noise dataset rather than clean data (ablation) | CombCls · CUB-200 open-set | accuracy (%): 77.82 -> 79.19 | Mixed |
| use CombCls feature extractor for a standard classifier (ablation) | Standard · CUB-200 open-set high uniform noise | accuracy (%): 70.04 -> 78.48 | Improved |
| use CombCls feature extractor for a standard classifier (ablation) | Standard · CUB-200 closed-set high uniform noise | accuracy (%): 42.35 -> 53.83 | Improved |
Earlier methods this paper compares with
Named baselines from earlier accepted papers in this paper's results tables: better than 1 of 1 on most shared results.
- Co-teaching — Co-teaching: Robust training of deep neural networks with extremely noisy labels (NeurIPS 2018) · better on 5 of 5 shared results
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
- Noise-Robust Learning from Multiple Unsupervised Sources of Inferred Labels (AAAI 2022 · similarity 0.88)
- Centrality and Consistency: Two-Stage Clean Samples Identification for Learning with Instance-Dependent Noisy Labels (ECCV 2022 · similarity 0.87)
- Deep Noise Label Learning via Effective Rank Reduction (ECCV 2026 · similarity 0.87)
- Estimating Noisy Class Posterior with Part-level Labels for Noisy Label Learning (CVPR 2024 · similarity 0.86)
- Foster Adaptivity and Balance in Learning with Noisy Labels (ECCV 2024 · similarity 0.86)
- Learning with Open-world Noisy Data via Class-independent Margin in Dual Representation Space (AAAI 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.
