Research map / Label-efficient and robust learning / Noisy labels
Robust Contrastive Learning Against Noisy Views
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 |
|---|---|---|---|
| Replace InfoNCE with RINCE | SimCLR and MoCov3 · ImageNet linear evaluation | Top 1 accuracy: 69.3 -> 70.0; 73.8 -> 74.2 | Helped |
| Replace InfoNCE with RINCE | Audio-visual contrastive learning · Kinetics400 pretraining; HMDB51 | Top-1 accuracy: 57.8 -> 61.6 | Helped |
| Replace InfoNCE with RINCE | Audio-visual contrastive learning · Kinetics400 pretraining; UCF101 | Top-1 accuracy: 88.6 -> 88.8 | Helped |
| Replace InfoNCE with RINCE | Graph contrastive learning · TUDataset RDT-B, NCI1, PROTEINS, and DD | accuracy: 89.9$\pm$0.4 -> 90.9$\pm$0.6; 78.2$\pm$0.8 -> 78.6$\pm$0.4; 74.4$\pm$0.5 -> 74.7$\pm$0.8; 78.6$\pm$0.8 -> 78.7$\pm$0.4 | Helped |
| Set RINCE q=1.0 rather than using InfoNCE (ablation) | SimCLR · CIFAR-10 label noise, η=0.8 | Top-1 accuracy: 87.1$\pm$0.5 -> 91.6$\pm$0.3 | Helped |
| Set RINCE q=1.0 rather than using InfoNCE (ablation) | SimCLR · CIFAR-10 augmentation noise, η=0.4 | Top-1 accuracy: 87.3$\pm$0.4 -> 89.0$\pm$0.1 | Helped |
| Use RINCE instead of InfoNCE | Audio-visual contrastive learning · UCF101, ACAV100M pretraining | Top-1 accuracy: 72.429 -> 74.253; 79.94 -> 82.368 | Helped |
| Vary q from 0.01 to 1.0 (ablation) | RINCE · CIFAR-10 label noise and augmentation noise | Top-1 accuracy: Label noise η=0.0: 93.4$7$0.2 -> 93.0$7$0.2; η=0.8: 89.1$7$0.2 -> 91.6$7$0.3; augmentation noise η=0.4: 87.7$7$0.5 | Mixed |
| Set λ between 0.01 and 0.1 (ablation) | RINCE · CIFAR-10 augmentation noise, η=0.4 | Top-1 accuracy: 89.65; 89.81; 89.9 | No clear effect |
Closest papers by meaning
- Conditional Negative Sampling for Contrastive Learning of Visual Representations (ICLR 2021 · similarity 0.91)
- Partially View-Aligned Representation Learning With Noise-Robust Contrastive Loss (CVPR 2021 · similarity 0.89)
- Investigating Why Contrastive Learning Benefits Robustness against Label Noise (ICML 2022 · similarity 0.89)
- Your contrastive learning problem is secretly a distribution alignment problem (NeurIPS 2024 · similarity 0.89)
- On the Similarities of Embeddings in Contrastive Learning (ICML 2025 · similarity 0.88)
- $\mathbb{X}$-Sample Contrastive Loss: Improving Contrastive Learning with Sample Similarity Graphs (ICLR 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.