Research map / Label-efficient and robust learning / Self-supervised and contrastive
On the Importance of Asymmetry for Siamese Representation Learning
Check 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 |
|---|---|---|---|
| Apply high-variance MultiCrop to the source encoder (ablation) | MoCo v2 · ImageNet linear probing, 100 epochs | top-1 accuracy (%): 65.8 -> 69.9 (+4.1%) | Helped |
| Apply high-variance MultiCrop to the target encoder (ablation) | MoCo v2 · ImageNet linear probing, 100 epochs | top-1 accuracy (%): 65.8 -> 57.1 (-8.7%) | Hurt |
| Apply high-variance ScaleMix to the source encoder (ablation) | MoCo v2 · ImageNet linear probing, 100 epochs | top-1 accuracy (%): 65.8 -> 67.3 (+1.5%) | Helped |
| Apply high-variance ScaleMix to the target encoder (ablation) | MoCo v2 · ImageNet linear probing, 100 epochs | top-1 accuracy (%): 65.8 -> 52.8 (-13.0%) | Hurt |
| Apply weaker augmentation to the target encoder (ablation) | MoCo v2 · ImageNet linear probing, 100 epochs | top-1 accuracy (%): 65.8 -> 67.2 (+1.4%) | Helped |
| Apply weaker augmentation to the source encoder (ablation) | MoCo v2 · ImageNet linear probing, 100 epochs | top-1 accuracy (%): 65.8 -> 51.0 (-14.8%) | Hurt |
| Apply stronger augmentation to the source encoder (ablation) | MoCo v2 · ImageNet linear probing, 100 epochs | top-1 accuracy (%): 65.8 -> 66.7 (+0.9%) | Helped |
| Apply stronger augmentation to the target encoder (ablation) | MoCo v2 · ImageNet linear probing, 100 epochs | top-1 accuracy (%): 65.8 -> 62.2 (-3.6%) | Hurt |
| Replace target projector BN with SyncBN to reduce target variance (ablation) | MoCo v2 · ImageNet linear probing, 100 epochs | top-1 accuracy (%): 65.8 -> 66.5 (+0.7%) | Helped |
| Replace source projector BN with SyncBN (ablation) | MoCo v2 · ImageNet linear probing, 100 epochs | top-1 accuracy (%): 65.8 -> 64.7 (-0.9%) | Hurt |
| Allocate more MeanEnc views to target (1 source, 3 target rather than 3 source, 1 target) (ablation) | MoCo v2 · ImageNet linear probing, 100 epochs | top-1 accuracy (%): 59.9 -> 67.9 | Helped |
| Combine source MultiCrop, target AsymBN, and target MeanEnc (ablation) | MoCo v2 · ImageNet linear probing, 100 epochs | top-1 accuracy (%): 65.8 -> 71.3 (+5.5%) | Helped |
Closest papers by meaning
- Exploring Simple Siamese Representation Learning (CVPR 2021 · similarity 0.90)
- Asymmetric Dual Self-Distillation for 3D Self-Supervised Representation Learning (NeurIPS 2025 · similarity 0.86)
- AsCAN: Asymmetric Convolution-Attention Networks for Efficient Recognition and Generation (NeurIPS 2024 · similarity 0.85)
- In-Context Symmetries: Self-Supervised Learning through Contextual World Models (NeurIPS 2024 · similarity 0.85)
- Rethinking Multi-view Representation Learning via Distilled Disentangling (CVPR 2024 · similarity 0.84)
- Asymmetric Duos: Sidekicks Improve Uncertainty (NeurIPS 2025 · similarity 0.84)
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.