Research map / Label-efficient and robust learning / Self-supervised and contrastive
On the Surrogate Gap between Contrastive and Supervised Losses
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 |
|---|---|---|---|
| Increase negative sample size K | Contrastive learning on the synthetic circle dataset · Synthetic circle test set | mean supervised loss | Helped |
| Increase negative sample size K | Contrastive learning on the synthetic circle dataset · Synthetic circle test set | variance of supervised performance | Helped |
| Increase negative sample size K | Contrastive learning with Wiki-3029 · Wiki-3029 | accuracy | Mixed |
| Increase negative sample size K | Contrastive learning with CIFAR-10/100 · CIFAR-10/100 | accuracy | Hurt |
| Increase negative sample size K | Contrastive learning with CIFAR-10/100 · CIFAR-10/100 | supervised loss | No clear effect |
| Train for more epochs instead of increasing K | Contrastive learning with small K and large C · CIFAR-100 | classification accuracy | Helped |
| Use the proposed surrogate bounds | Contrastive-to-supervised loss estimation · CIFAR-10 | bound tightness | Helped |
| Increase negative sample size K | The surrogate gap between contrastive and mean supervised losses · All K under the stated assumptions | surrogate gap: O(K^{-1}) | Helped |
Closest papers by meaning
- $\mathbb{X}$-Sample Contrastive Loss: Improving Contrastive Learning with Sample Similarity Graphs (ICLR 2025 · similarity 0.89)
- On the Similarities of Embeddings in Contrastive Learning (ICML 2025 · similarity 0.88)
- How does Labeling Error Impact Contrastive Learning? A Perspective from Data Dimensionality Reduction (ICML 2025 · similarity 0.88)
- Diversity Is All You Need for Contrastive Learning: Spectral Bounds on Gradient Magnitudes (NeurIPS 2025 · similarity 0.88)
- Self-Supervised Contrastive Learning is Approximately Supervised Contrastive Learning (NeurIPS 2025 · similarity 0.87)
- An Investigation of Representation and Allocation Harms in Contrastive Learning (ICLR 2024 · 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.