Research map / Label-efficient and robust learning / Recognition and long-tail
Understanding Contrastive Learning via Distributionally Robust Optimization
Junkang Wu, Jiawei Chen, Jiancan Wu, Wentao Shi, Xiang Wang, Xiangnan He
Read the paperCheck your idea against 137,217 papers
AI-extracted. The measured effects below were read by AI from the paper's full text (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 ADNCE | SimCLR, ResNet50 · CIFAR10, 400 epochs | Top-1 accuracy: 91.64 -> 91.88 | Improved |
| Replace InfoNCE with ADNCE | SimCLR, ResNet50 · STL10, 400 epochs | Top-1 accuracy: 87.69 -> 88.00 | Improved |
| Replace InfoNCE with ADNCE | SimCLR, ResNet50 · CIFAR100, 400 epochs | Top-1 accuracy: 69.03 -> 69.35 | Improved |
| Replace InfoNCE with ADNCE | SimCSE-BERTbase · STS average | Spearman correlation: 76.81 -> 77.03 | Improved |
| Replace InfoNCE with ADNCE | SimCSE-RoBERTabase · STS average | Spearman correlation: 76.81 -> 77.10 | Improved |
| Tune temperature τ by grid search instead of using τ₀ | SimCLR, ResNet50 · STL10, 400 epochs | Top-1 accuracy: 81.67 -> 87.69 | Improved |
| Tune temperature τ by grid search instead of using τ₀ | SimCLR, ResNet50 · CIFAR100, 400 epochs | Top-1 accuracy: 66.53 -> 69.03 | Improved |
| Tune temperature τ by grid search instead of using τ₀ | SimCSE-BERTbase · STS average | Spearman correlation: 76.25 -> 76.81 | Improved |
| Tune temperature τ by grid search instead of using τ₀ | SimCSE-RoBERTabase · STS average | Spearman correlation: 76.57 -> 76.81 | Improved |
| Replace InfoNCE with the mean-variance objective (ablation) | Contrastive learning model · Reported benchmark experiments | Performance | Hurt |
Closest papers by meaning
- Model-Aware Contrastive Learning: Towards Escaping the Dilemmas (ICML 2023 · similarity 0.89)
- Your contrastive learning problem is secretly a distribution alignment problem (NeurIPS 2024 · similarity 0.89)
- Do More Negative Samples Necessarily Hurt In Contrastive Learning? (ICML 2022 · similarity 0.88)
- Aligning Visual Contrastive learning models via Preference Optimization (ICLR 2025 · similarity 0.88)
- On the Similarities of Embeddings in Contrastive Learning (ICML 2025 · similarity 0.87)
- A Statistical Theory of Contrastive Learning via Approximate Sufficient Statistics (NeurIPS 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.
