Research map / Label-efficient and robust learning / Self-supervised and contrastive
What Should Not Be Contrastive in 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 |
|---|---|---|---|
| Add rotation augmentation to standard MoCo | MoCo · IN-100 rotation classification | accuracy: 61.1 -> 43.3 | Hurt |
| Add rotation augmentation to standard MoCo | MoCo · IN-100 | top-1 accuracy: 81.0 -> 79.4 | Hurt |
| Use LooC rather than MoCo with color augmentation | MoCo · iNat-1k | top-1 accuracy: 36.2 -> 41.2 | Helped |
| Add rotation-dependent representation alongside color (ablation) | LooC with color · Flowers-102 | 5-shot accuracy: 68.2 ($\pm$ 0.6) -> 70.9 ($\pm$ 0.3) | Helped |
| Add rotation-dependent representation alongside color (ablation) | LooC with color · CUB-200 | top-1 accuracy: 40.1 -> 39.6 | Hurt |
| Concatenate augmentation-specific embedding spaces | LooC · iNat-1k | top-1 accuracy: 44.0 -> 46.1 | Helped |
| Concatenate augmentation-specific embedding spaces | LooC · IN-100 | top-1 accuracy: 79.2 -> 81.2 | Helped |
| Use rotation augmentation in LooC | LooC · ON-13 | top-1 accuracy: 29.2 -> 34.2 | Helped |
| Use texture randomization in LooC | LooC · IN-C-100 | all top-1 accuracy: 48.2 -> 51.3 | Helped |
| Add texture-dependent features to the concatenated representation (ablation) | LooC++ · IN-C-100 | all top-1 accuracy: 52.6 -> 44.1 | Hurt |
| Add rotation-dependent features to the concatenated representation (ablation) | LooC++ · IN-C-100 | all top-1 accuracy: 44.1 -> 48.0 | Helped |
| Replace leave-one-out views with add-one views (ablation) | MoCo · IN-100 | top-1 accuracy: 81.0 -> 74.9 | Hurt |
Closest papers by meaning
- Intriguing Properties of Contrastive Losses (NeurIPS 2021 · similarity 0.91)
- A Simple Framework for Contrastive Learning of Visual Representations (ICML 2020 · similarity 0.91)
- Improving Transformation Invariance in Contrastive Representation Learning (ICLR 2021 · similarity 0.90)
- Variational Supervised Contrastive Learning (NeurIPS 2025 · similarity 0.90)
- $\mathbb{X}$-Sample Contrastive Loss: Improving Contrastive Learning with Sample Similarity Graphs (ICLR 2025 · similarity 0.88)
- A Statistical Theory of Contrastive Learning via Approximate Sufficient Statistics (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.