Research map / Label-efficient and robust learning / Semi-supervised learning
Rethinking Weak Supervision in Helping Contrastive Learning
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 clean semi-supervised labels jointly with unsupervised contrastive learning | Jointly trained contrastive learning · Semi-supervised setting | downstream linear evaluation error bound | Helped |
| Jointly train supervised and unsupervised contrastive losses using noisy labels | SimCLR and SupCon · Symmetric label noise | downstream linear evaluation error bound | No clear effect |
| Choose the better of SupCon and SimCLR instead of relying on their joint training | Noisy-label contrastive learning · CIFAR-10 and TinyImageNet-200 | linear probing accuracy | No clear effect |
| Increase the SupCon weight θ (ablation) | JointTraining (SimCLR and SupCon) · TinyImageNet-200, noise rates 0.4 and 0.8 | linear probing accuracy | Hurt |
| Increase feature dimension k from 128 to 256 (ablation) | JointTraining · CIFAR-10, 30% label noise | linear probing accuracy: 91.89 -> 92.18 | Helped |
| Increase feature dimension k beyond 256 (ablation) | JointTraining · CIFAR-10, 30% label noise | linear probing accuracy: 92.18 -> 92.12 -> 91.77 -> 90.9 | Hurt |
| Increase the noise rate while using joint training | JointTraining (SimCLR and SupCon) · Noisy-label contrastive learning | linear probing accuracy | Mixed |
Closest papers by meaning
- Weakly-Supervised Contrastive Learning for Imprecise Class Labels (ICML 2025 · similarity 0.91)
- Learning from weak labelers as constraints (ICLR 2025 · similarity 0.89)
- Self-Supervised Contrastive Learning is Approximately Supervised Contrastive Learning (NeurIPS 2025 · similarity 0.88)
- How does Labeling Error Impact Contrastive Learning? A Perspective from Data Dimensionality Reduction (ICML 2025 · similarity 0.88)
- Continuous Contrastive Learning for Long-Tailed Semi-Supervised Recognition (NeurIPS 2024 · similarity 0.87)
- On the Similarities of Embeddings in Contrastive Learning (ICML 2025 · similarity 0.87)
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.