Research map / Label-efficient and robust learning / OOD and anomaly detection
Neural Relation Graph: A Unified Framework for Identifying Label Noise and Outlier Data
Jang-Hyun Kim, Sangdoo Yun, Hyun Oh Song
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 |
|---|---|---|---|
| Use relational graph scores instead of the strongest baseline, Margin | Label error detection · ImageNet, 8% label noise, MAE-Large | AP: 0.484 -> 0.526 | Improved |
| Use relational graph scores instead of the strongest baseline, TracIn | Label error detection · ImageNet, 8% label noise, MAE-Large | TNR95: 0.521 -> 0.695 | Improved |
| Use relational graph scores instead of the strongest baseline | Label error detection · ESC-50 | AP: 0.739 -> 0.779 | Improved |
| Use relational graph scores instead of the strongest baseline | Label error detection · SST2 | AP: 0.861 -> 0.881 | Improved |
| Use relational graph scores instead of the strongest baseline | Label error detection · MNLI | AP: 0.764 -> 0.758 | Hurt |
| Use relational graph scores instead of the strongest baseline | Label error detection · MNLI | TNR95: 0.514 -> 0.660 | Improved |
| Use relational graph scores instead of the strongest baseline | Realistic label error detection · ImageNet validation with ConvNeXt-22k | AP: 0.724 -> 0.744 | Improved |
| Use relational graph scores instead of KNN | OOD detection · ImageNet OOD, ALL, MAE-Large | AUROC: 0.901 -> 0.911 | Improved |
| Use relational graph scores instead of KNN | OOD detection · ImageNet OOD, ALL, MAE-Large | AP: 0.854 -> 0.874 | Improved |
| Use relational scores instead of KNN | Outlier detection · ImageNet-100 with SUN outliers | AUROC: 0.990 -> 0.993 | Improved |
| Use relational scores instead of KNN | Outlier detection · ImageNet-100 with iNaturalist outliers | AP: 0.923 -> 0.940 | Improved |
| Add max-cut global edge aggregation rather than simply summing node edges (ablation) | Relation label-error scoring · Label error detection | TNR95: 0.628 -> 0.647 | Improved |
Closest papers by meaning
- DualGraph: A Graph-Based Method for Reasoning About Label Noise (CVPR 2021 · similarity 0.88)
- Learning From Noisy Labels With Distillation (ICCV 2017 · similarity 0.87)
- Detecting Corrupted Labels Without Training a Model to Predict (ICML 2022 · similarity 0.87)
- ML-GOOD: Towards Multi-Label Graph Out-Of-Distribution Detection (AAAI 2025 · similarity 0.86)
- Noisy Label Learning with Instance-Dependent Outliers: Identifiability via Crowd Wisdom (NeurIPS 2024 · similarity 0.86)
- Learning with Open-world Noisy Data via Class-independent Margin in Dual Representation Space (AAAI 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.
