Research map / Label-efficient and robust learning / Noisy labels
Iterative Learning With Open-Set Noisy Labels
Read the paperCheck your idea against 108,967 papers
AI-extracted summary. Written by AI from the paper's title only (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 the iterative noisy-label detection, contrastive feature learning, and reweighting framework | CNN training with open-set label noise · CIFAR-10 + CIFAR-100/SVHN and other 40% open-set noise settings | accuracy: 79.28 vs 64.58; 79.38 vs 67.53; 77.73 vs 56.93 | Helped |
| use the proposed framework | CNN training with closed-set label noise · CIFAR-10, 20% closed-set noise | accuracy: 81.36 vs 80.15 | Helped |
| use the proposed framework | CNN training with closed-set label noise · CIFAR-10, 40% closed-set noise | accuracy: 78.15 vs 78.69 | Hurt |
| use the proposed framework instead of CNN-CRF | ResNet-50 and Inception-v3 · ImageNet 200 clean classes with 20% open-set noise | Top-1 accuracy: 70.29 vs 66.54; 71.43 vs 67.23 | Helped |
| use the proposed framework instead of CNN-CRF | ResNet-50 and Inception-v3 feature learning · CIFAR-100 test using web-search training data | accuracy: 67.90 vs 63.94; 64.21 vs 60.47 | Helped |
| remove adaptive reweighting by setting γ=1 (ablation) | Ours · CIFAR-10 + CIFAR-100, 20% and 40% open-set noise | accuracy: 81.96 -> 76.97; 79.28 -> 74.45 | Hurt |
| ignore detected noisy samples by setting γ=0 (ablation) | Ours · CIFAR-10 + CIFAR-100, 20% and 40% open-set noise | accuracy: 81.96 -> 79.27; 79.28 -> 76.03 | Hurt |
| remove detected noisy samples and omit contrastive feature learning (ablation) | Ours · CIFAR-10 + CIFAR-100, 20% and 40% open-set noise | accuracy: 81.96 -> 76.22; 79.28 -> 68.40 | Hurt |
| label detected noisy samples as a new unknown class rather than use contrastive learning (ablation) | Ours · CIFAR-10 + CIFAR-100, 20% and 40% open-set noise | accuracy: 81.96 -> 78.34; 79.28 -> 73.11 | Hurt |
| perform noisy-label detection only once instead of iteratively (ablation) | Ours · CIFAR-10 + CIFAR-100, 20% and 40% open-set noise | accuracy: 81.96 -> 77.52; 79.28 -> 70.31 | Hurt |
| remove noisy-label detection and its softmax reweighting (ablation) | Ours · CIFAR-10 + CIFAR-100, 20% and 40% open-set noise | accuracy: 81.96 -> 76.17; 79.28 -> 63.50 | Hurt |
| vary contrastive-loss weight η from 0.5 to 1.5 (ablation) | Ours · CIFAR-10 + CIFAR-100, 40% open-set noise | accuracy: 76.08; 79.04; 79.36; 79.77; 79.08; 77.80 | No clear effect |
Closest papers by meaning
- Open-set Label Noise Can Improve Robustness Against Inherent Label Noise (NeurIPS 2021 · similarity 0.90)
- Toward Robustness against Label Noise in Training Deep Discriminative Neural Networks (NeurIPS 2017 · similarity 0.90)
- Unlocking the Power of Open Set: A New Perspective for Open-Set Noisy Label Learning (AAAI 2024 · similarity 0.89)
- Learning with Structural Labels for Learning with Noisy Labels (CVPR 2024 · similarity 0.89)
- Learning with Open-world Noisy Data via Class-independent Margin in Dual Representation Space (AAAI 2025 · similarity 0.88)
- On the Role of Label Noise in the Feature Learning Process (ICML 2025 · similarity 0.88)
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.