Research map / Label-efficient and robust learning / Noisy labels
Investigating Why Contrastive Learning Benefits Robustness against Label Noise
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 |
|---|---|---|---|
| Add SimCLR pretraining before ELR fine-tuning | ELR with ResNet-32 · CIFAR-10, Sym 80 | test accuracy: 69.5 \pm 5.0 -> 77.7 \pm 0.2 | Helped |
| Add SimCLR pretraining before Crust fine-tuning | Crust with ResNet-32 · CIFAR-10, Sym 80 | test accuracy: 33.8 \pm {1.3} -> 77.9 \pm 0.3 | Helped |
| Add SimCLR pretraining before Mixup fine-tuning | Mixup with ResNet-32 · CIFAR-10, Sym 80 | test accuracy: 40.7 \pm {1.1} -> 74.8 \pm 0.3 | Helped |
| Add SimCLR pretraining before ELR fine-tuning | ELR with ResNet-32 · CIFAR-100, Sym 80 | test accuracy: 23.4 \pm {1.9} -> 35.9 \pm 0.3 | Helped |
| Add SimCLR pretraining before end-to-end fine-tuning | ResNet-32 end-to-end fine-tuning · CIFAR-100, Sym 80 | test accuracy: 15.1 \pm {0.1} -> 36.7 \pm 0.3 | Helped |
| Add SimCLR pretraining before ELR fine-tuning | ELR with InceptionResNet-v2 · WebVision | Top-1 accuracy: 76.26 -> 79.52 | Helped |
| Add SimCLR pretraining before Mixup fine-tuning | Mixup with InceptionResNet-v2 · ImageNet | Top-1 accuracy: 68.34 -> 72.25 | Helped |
| Increase network size from ResNet-32 to ResNet-34 (ablation) | CL-pretrained network fine-tuned with Mixup · CIFAR-10, Sym 80 | test accuracy: 74.8 $\pm$.3 -> 90.8$\pm$.6 | Helped |
| Increase network size from ResNet-32 to ResNet-34 (ablation) | CL-pretrained network fine-tuned with Mixup · CIFAR-100, Sym 80 | test accuracy: 30.3$\pm$.4 -> 69.4$\pm$.3 | Helped |
| Increase network size from ResNet-32 to ResNet-34 (ablation) | CL-pretrained network fine-tuned with Mixup · CIFAR-10, Asym 40 | test accuracy: 87.5$\pm$1.3 -> 90.4$\pm$.4 | Helped |
| Increase network size from ResNet-32 to ResNet-34 (ablation) | CL-pretrained network fine-tuned with Mixup · CIFAR-100, Asym 40 | test accuracy: 55.5$\pm$.8 -> 65.2$\pm$.2 | Helped |
Closest papers by meaning
- Selective-Supervised Contrastive Learning With Noisy Labels (CVPR 2022 · similarity 0.90)
- When does Contrastive Learning Preserve Adversarial Robustness from Pretraining to Finetuning? (NeurIPS 2021 · similarity 0.90)
- On Learning Contrastive Representations for Learning With Noisy Labels (CVPR 2022 · similarity 0.90)
- Robust Pre-Training by Adversarial Contrastive Learning (NeurIPS 2020 · similarity 0.89)
- How does Labeling Error Impact Contrastive Learning? A Perspective from Data Dimensionality Reduction (ICML 2025 · similarity 0.89)
- Self-Supervised Contrastive Learning is Approximately Supervised Contrastive Learning (NeurIPS 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.