Research map / Label-efficient and robust learning / Recognition and long-tail
Learning Contrastive Embedding in Low-Dimensional Space
Shuo Chen, Chen Gong, Jun Li, Jian Yang, Gang Niu, Masashi Sugiyama
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 nuclear-norm low-rank reconstruction (256-dim) vs no regularizer (ablation) | SimCLR · STL-10 400 epochs | accuracy (%): 74.1 -> 79.2 | Improved |
| add nuclear-norm low-rank reconstruction (256-dim) vs no regularizer (ablation) | SimCLR · CIFAR-10 400 epochs | accuracy (%): 88.4 -> 93.1 | Improved |
| add l2,1-norm regularizer to 512-dim embedding (ablation) | SimCLR · CIFAR-10 400 epochs | accuracy (%): 90.1 -> 92.5 | Improved |
| increase embedding dimension from 512 to 4096 without regularizer (ablation) | SimCLR · CIFAR-10 400 epochs | accuracy (%): 90.3 -> 85.4 | Hurt |
| reduce embedding dimension naively from 2048 to 256 without regularizer (ablation) | SimCLR · STL-10 400 epochs | accuracy (%): 76.2 -> 74.2 | No clear effect |
| low-rank regularizer at early training (100 epochs) (ablation) | CLLR · STL-10 100 epochs | accuracy (%): 56.4 -> 56.3 | No clear effect |
| low-rank regularizer at early training (100 epochs) (ablation) | CLLR · CIFAR-10 100 epochs | accuracy (%): 66.4 -> 67.5 | No clear effect |
| add CLLR (nuclear-norm) to DCL | DCL · BookCorpus MPQA | accuracy (%): 93.7 -> 95.8 | Improved |
| add CLLR (nuclear-norm) to DCL | DCL · BookCorpus TREC | accuracy (%): 89.1 -> 91.2 | Improved |
| add CLLR to CMC | CMC · ImageNet-100 neg. 4096 | Top-1 accuracy (%): 73.58 -> 76.91 | Improved |
| add CLLR (nuclear-norm) to HCL | HCL · ImageNet-100 neg. 1024 | Top-1 accuracy (%): 60.89 -> 62.43 | Improved |
| add CLLR (nuclear-norm) to CMC at small negative count | CMC · ImageNet-100 neg. 1024 | Top-1 accuracy (%): 60.23 -> 61.23 | Improved |
Closest papers by meaning
- Contrastive Learning with the Feature Reconstruction Amplifier (AAAI 2023 · similarity 0.90)
- Which Features are Learnt by Contrastive Learning? On the Role of Simplicity Bias in Class Collapse and Feature Suppression (ICML 2023 · similarity 0.88)
- On the Similarities of Embeddings in Contrastive Learning (ICML 2025 · similarity 0.87)
- Variational Supervised Contrastive Learning (NeurIPS 2025 · similarity 0.86)
- Embedding Dimension of Contrastive Learning and $k$-Nearest Neighbors (NeurIPS 2024 · similarity 0.86)
- Generalization Analysis for Supervised Contrastive Representation Learning under Non-IID Settings (ICML 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.
