Research map / Label-efficient and robust learning / Self-supervised and contrastive
Towards the Generalization of Contrastive Self-Supervised 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 the omitted transformations successively, from cropping only through all five transformations (ablation) | SimCLR, Barlow Twins, MoCo, and SimSiam · CIFAR-10 | accuracy: 62.74 ± 0.18 -> 89.76 ± 0.12 (SimCLR); 67.77 ± 0.69 -> 86.91 ± 0.09 (Barlow Twins); 74.94 ± 0.22 -> 90.12 ± 0.12 (MoCo); | Helped |
| Add the omitted transformations successively, from cropping only through all five transformations (ablation) | SimCLR, Barlow Twins, MoCo, and SimSiam · CIFAR-100 | accuracy: 27.95 ± 0.09 -> 57.74 ± 0.12 (SimCLR); 34.05 ± 1.13 -> 57.99 ± 0.29 (Barlow Twins); 39.63 ± 0.31 -> 64.19 ± 0.14 (MoCo); | Helped |
| Use color distortion strength 1 instead of 1/8 (ablation) | SimCLR, Barlow Twins, MoCo, and SimSiam · CIFAR-10 | accuracy: 74.23 ± 0.16 -> 82.75 ± 0.24 (SimCLR); 77.96 ± 0.16 -> 82.58 ± 0.25 (Barlow Twins); 81.20 ± 0.12 -> 86.68 ± 0.05 (MoCo); | Helped |
| Use color distortion strength 1 instead of 1/8 (ablation) | SimCLR, Barlow Twins, MoCo, and SimSiam · CIFAR-100 | accuracy: 34.75 ± 0.20 -> 46.67 ± 0.42 (SimCLR); 44.72 ± 0.26 -> 50.39 ± 1.09 (Barlow Twins); 49.43 ± 0.16 -> 58.50 ± 0.51 (MoCo); | Helped |
| Include color dropping and color distortion in the augmentation set (ablation) | SimCLR, Barlow Twins, MoCo, and SimSiam · CIFAR-10 and CIFAR-100 | accuracy | Helped |
| Use a two-transformation pair rather than the other pairs, with random cropping fixed (ablation) | SimCLR · CIFAR-10 | classification error rate | Mixed |
Closest papers by meaning
- An Augmentation-Aware Theory for Self-Supervised Contrastive Learning (ICML 2025 · similarity 0.91)
- Chaos is a Ladder: A New Theoretical Understanding of Contrastive Learning via Augmentation Overlap (ICLR 2022 · similarity 0.91)
- ArCL: Enhancing Contrastive Learning with Augmentation-Robust Representations (ICLR 2023 · similarity 0.90)
- Self-Supervised Contrastive Learning is Approximately Supervised Contrastive Learning (NeurIPS 2025 · similarity 0.90)
- Generalization Analysis for Supervised Contrastive Representation Learning under Non-IID Settings (ICML 2025 · similarity 0.89)
- 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.