Research map / Label-efficient and robust learning / Recognition and long-tail
Embedding Transfer With Label Relaxation for Improved Metric 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 |
|---|---|---|---|
| Add source-space similarities as relaxed labels instead of binary class labels (ablation) | Contrastive metric learning · CUB-200-2011 | Recall@1: 65.3 -> 70.4 | Helped |
| Add source-space similarities as relaxed labels instead of binary class labels (ablation) | Contrastive metric learning · Cars-196 | Recall@1: 80.5 -> 88.2 | Helped |
| Use relative pairwise distances rather than absolute distances (ablation) | Relaxed contrastive loss · CUB-200-2011 | Recall@1: 70.4 -> 72.1 | Helped |
| Use relative pairwise distances rather than absolute distances (ablation) | Relaxed contrastive loss · Cars-196 | Recall@1: 88.2 -> 89.6 | Helped |
| Add multi-view data augmentation (ablation) | Relaxed contrastive transfer · CUB-200-2011 | Recall@1: 71.5 -> 72.1 | Helped |
| Add multi-view data augmentation (ablation) | Relaxed contrastive transfer · Cars-196 | Recall@1: 89.0 -> 89.6 | Helped |
| Transfer with relaxed MS loss instead of relaxed contrastive loss (ablation) | Embedding transfer · Self-transfer across CUB-200-2011, Cars-196, and SOP | Recall@1: 72.1 -> 72.3; 89.6 -> 89.2; 79.8 -> 79.3 | Mixed |
| Transfer with relaxed MS loss instead of relaxed contrastive loss (ablation) | Embedding transfer with 64D target · Dimensionality reduction across CUB-200-2011, Cars-196, and SOP | Recall@1: 67.4 -> 67.5; 86.5 -> 86.0; 76.3 -> 75.4 | Mixed |
| Use relaxed contrastive transfer instead of no embedding transfer | SimCLR representation · CIFAR-10 | linear evaluation accuracy: 93.4 -> 93.9 | Helped |
| Use relaxed contrastive transfer instead of no embedding transfer | SimCLR representation · STL-10 | linear evaluation accuracy: 89.2 -> 89.6 | Helped |
| Add relaxed contrastive transfer alongside HKD | ResNet56 → ResNet20 · CIFAR-100 | test accuracy: 71.63 -> 71.95 | Helped |
| Add relaxed contrastive transfer alongside HKD | VGG13 → VGG8 · CIFAR-100 | test accuracy: 74.29 -> 73.82 | Hurt |
Closest papers by meaning
- Self-Taught Metric Learning Without Labels (CVPR 2022 · similarity 0.89)
- Adapted Deep Embeddings: A Synthesis of Methods for k-Shot Inductive Transfer Learning (NeurIPS 2018 · similarity 0.88)
- On the Similarities of Embeddings in Contrastive Learning (ICML 2025 · similarity 0.87)
- Joint‑Embedding vs Reconstruction: Provable Benefits of Latent Space Prediction for Self‑Supervised Learning (NeurIPS 2025 · similarity 0.86)
- Fully Dynamic Embedding into $\ell_p$ Spaces (ICML 2025 · similarity 0.86)
- $\mathbb{X}$-Sample Contrastive Loss: Improving Contrastive Learning with Sample Similarity Graphs (ICLR 2025 · similarity 0.85)
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.