Research map / Label-efficient and robust learning / Noisy labels
Selective-Supervised Contrastive Learning With Noisy Labels
Check 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 |
|---|---|---|---|
| Select confident pairs from noisy positive pairs using representation similarity (ablation) | Sel-CL · CIFAR-100 asymmetric 40% noise | weighted KNN accuracy: 70.25 -> 72.71 | Helped |
| Add Mixup data augmentation (ablation) | Sel-CL · CIFAR-100 asymmetric 40% noise | test accuracy: 64.2 -> 72.0 | Helped |
| Add MOCO trick (ablation) | Sel-CL · CIFAR-100 asymmetric 40% noise | test accuracy: 69.2 -> 72.0 | Helped |
| Select confident pairs rather than use all examples and all pairs (ablation) | Sel-CL · CIFAR-100 asymmetric 40% noise | weighted KNN accuracy: 68.66 -> 72.71 | Helped |
| Add classifier learning on confident examples (ablation) | Sel-CL · CIFAR-100 asymmetric 40% noise | weighted KNN accuracy: 70.2 -> 72.7 | Helped |
| Add similarity loss (ablation) | Sel-CL · CIFAR-100 asymmetric 40% noise | weighted KNN accuracy: 72.5 -> 72.7 | Helped |
| Use weak rather than strong augmentation during fine-tuning (ablation) | Sel-CL+ · CIFAR-100 asymmetric 40% noise | test accuracy: 72.7 -> 74.2 | No clear effect |
| Retrain the classifier during fine-tuning (ablation) | Sel-CL+ · CIFAR-100 asymmetric 40% noise | test accuracy: 73.4 -> 74.2 | Helped |
| Use representation-similarity pseudo-labels to identify confident examples (ablation) | Sel-CL+ · CIFAR-100 asymmetric 40% noise | test accuracy: 69.4 -> 74.2 | Helped |
| Use Uns-CL rather than Sup-CL for warm-up (ablation) | Sel-CL+ · CIFAR-100 symmetric 90% noise | test accuracy: 51.4 -> 48.8 | Hurt |
| Use Sel-CL rather than Uns-CL for representation initialization (ablation) | ELR+ · CIFAR-100 asymmetric 40% noise | test accuracy: 72.3 -> 72.9 | Helped |
| Use Sel-CL rather than Uns-CL for representation initialization (ablation) | DivideMix · CIFAR-100 symmetric 20% noise | test accuracy: 78.3 -> 78.7 | Helped |
Closest papers by meaning
- Investigating Why Contrastive Learning Benefits Robustness against Label Noise (ICML 2022 · similarity 0.90)
- On Learning Contrastive Representations for Learning With Noisy Labels (CVPR 2022 · similarity 0.90)
- Fine-Grained Classification With Noisy Labels (CVPR 2023 · similarity 0.89)
- Twin Contrastive Learning With Noisy Labels (CVPR 2023 · similarity 0.88)
- On the Similarities of Embeddings in Contrastive Learning (ICML 2025 · similarity 0.88)
- Your contrastive learning problem is secretly a distribution alignment problem (NeurIPS 2024 · 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.