Research map / Detection and segmentation / Open-set and few-shot detection
A Low Rank Promoting Prior for Unsupervised Contrastive Learning
Yu Wang, Jingyang Lin, Qi Cai, Yingwei Pan, Ting Yao, Hongyang Chao, Tao Mei
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 |
|---|---|---|---|
| add low rank prior h(Q) (LORAC) over MoCo-M (ablation) | multi-view MoCo · ImageNet100 (ResNet-18) | top-1 accuracy: 77.0 -> 78.7 | Improved |
| add low rank prior h(Q) (LORAC) over MoCo-M (ablation) | multi-view MoCo · ImageNet1K linear (ResNet-50, 200 epochs) | top-1 accuracy: 72.5 -> 73.2 | Improved |
| replace nuclear norm with constant c in h(Q) (ablation) | LORAC · ImageNet100 (ResNet-18) | top-1 accuracy: 78.7 -> 76.8 | Hurt |
| use Gaussian prior instead of Laplace prior (ablation) | LORAC · ImageNet100 (ResNet-18) | top-1 accuracy: 78.7 -> 78.8 | No clear effect |
| use batch-shared P matrix (LORAC-bs) (ablation) | LORAC · ImageNet100 (ResNet-18) | top-1 accuracy: 78.7 -> 78.6 | No clear effect |
| add multiple views and multi-view MoCo over single-view MoCo-v2 | MoCo-v2 · ImageNet100 (ResNet-18) | top-1 accuracy: 65.3 -> 77.0 | Improved |
| use LORAC instead of SwAV multi-crop | SwAV · ImageNet1K linear (ResNet-50, 200 epochs) | top-1 accuracy: 72.7 -> 73.2 | Improved |
| add low rank prior (LORAC over MoCo-M) (ablation) | semi-supervised ResNet-50 · ImageNet 1% labels | top-1 accuracy: 48.4 -> 50.0 | Improved |
| add low rank prior (LORAC over MoCo-M) (ablation) | COCO detection ResNet-50 · COCO object detection | AP: 39.3 -> 39.5 | Improved |
| add low rank prior (LORAC over MoCo-M) (ablation) | transfer linear classification · 12 transfer datasets average | accuracy: 71.2 -> 72.4 | Improved |
| increase beta (weaken low rank prior) (ablation) | LORAC · ImageNet100 (ResNet-18) | top-1 accuracy | Hurt |
| increase number of views M (ablation) | LORAC · ImageNet100 (ResNet-18) | top-1 accuracy | Improved |
Closest papers by meaning
- Integrating Prior Knowledge in Contrastive Learning with Kernel (ICML 2023 · similarity 0.88)
- Conditional Negative Sampling for Contrastive Learning of Visual Representations (ICLR 2021 · similarity 0.87)
- AUC-CL: A Batchsize-Robust Framework for Self-Supervised Contrastive Representation Learning (ICLR 2024 · similarity 0.87)
- Prototypical Contrastive Learning of Unsupervised Representations (ICLR 2021 · similarity 0.87)
- BECLR: Batch Enhanced Contrastive Few-Shot Learning (ICLR 2024 · similarity 0.86)
- Variational Supervised Contrastive Learning (NeurIPS 2025 · similarity 0.86)
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.
