Research map / Label-efficient and robust learning / Noisy labels
Jo-SNC: Combating Noisy Labels Through Fostering Self- and Neighbor-Consistency
Zeren Sun, Yazhou Yao, Tongliang Liu, Zechao Li, Fumin Shen, JinHui TANG
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 |
|---|---|---|---|
| combine self-consistency and neighbor-consistency with divergence in selection (SelfNeiDiv) rather than self only (ablation) | Jo-SNC · CIFAR80N-O Sym-50% | test accuracy: 60.92 -> 62.70 | Improved |
| add neighbor consistency to selection rather than neighbors only (ablation) | Jo-SNC · CIFAR80N-O Sym-50% | test accuracy: 60.40 -> 62.70 | Improved |
| add selection criterion combination vs Dis only (ablation) | Jo-SNC · CIFAR80N-O Asym-40% | test accuracy: 62.65 -> 63.06 | Improved |
| use per-class dynamic threshold instead of pre-defined (ablation) | Jo-SNC · CIFAR80N-O Sym-50% | test accuracy: 59.91 -> 62.70 | Improved |
| use per-class dynamic threshold instead of pre-defined (ablation) | Jo-SNC · CIFAR80N-O Asym-40% | test accuracy: 57.74 -> 63.06 | Improved |
| use per-class threshold instead of GMM (ablation) | Jo-SNC · CIFAR80N-O Asym-40% | test accuracy: 59.81 -> 63.06 | Improved |
| use per-class threshold instead of mean threshold (ablation) | Jo-SNC · CIFAR80N-O Sym-50% | test accuracy: 60.92 -> 62.70 | Improved |
| train ID noisy samples with partial label learning rather than dropping them (ablation) | Jo-SNC · CIFAR80N-O Sym-50% | test accuracy: 59.55 -> 62.70 | Improved |
| train ID noisy samples with PLL rather than semi-supervised learning (ablation) | Jo-SNC · CIFAR80N-O Sym-50% | test accuracy: 61.84 -> 62.70 | Improved |
| train ID noisy samples with PLL rather than negative learning (ablation) | Jo-SNC · CIFAR80N-O Asym-40% | test accuracy: 62.12 -> 63.06 | Improved |
| Jo-SNC vs best prior method CA2C | Jo-SNC · Web-Aircraft/Bird/Car average | test accuracy: 86.43 -> 87.10 | Improved |
| Jo-SNC vs best prior method on Sym-80% CIFAR100N-C | Jo-SNC · CIFAR100N-C Sym-80% | test accuracy: 35.23 -> 43.45 | Improved |
Closest papers by meaning
- Jo-SRC: A Contrastive Approach for Combating Noisy Labels (CVPR 2021 · similarity 0.92)
- Neighborhood Collective Estimation for Noisy Label Identification and Correction (ECCV 2022 · similarity 0.91)
- Neighborhood-Regularized Self-Training for Learning with Few Labels (AAAI 2023 · similarity 0.90)
- Learning With Neighbor Consistency for Noisy Labels (CVPR 2022 · similarity 0.90)
- Combating Noisy Labels by Agreement: A Joint Training Method with Co-Regularization (CVPR 2020 · similarity 0.89)
- Centrality and Consistency: Two-Stage Clean Samples Identification for Learning with Instance-Dependent Noisy Labels (ECCV 2022 · similarity 0.89)
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.
