Research map / Label-efficient and robust learning / Zero-/few-shot and meta-learning
A Universal Representation Transformer Layer for Few-Shot Image Classification
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 |
|---|---|---|---|
| Replace SUR's hand-designed episode-time backbone weighting with meta-trained URT attention | Universal representations on Meta-Dataset · Omniglot | accuracy: 93.1$\pm$0.5 -> 94.4$\pm$0.4 | Helped |
| Replace SUR's hand-designed episode-time backbone weighting with meta-trained URT attention | Universal representations on Meta-Dataset · Aircraft | accuracy: 85.4$\pm$0.7 -> 85.8$\pm$0.6 | Helped |
| Replace SUR's hand-designed episode-time backbone weighting with meta-trained URT attention | Universal representations on Meta-Dataset · Birds | accuracy: 71.4$\pm$1.0 -> 76.3$\pm$0.8 | Helped |
| Replace SUR's hand-designed episode-time backbone weighting with meta-trained URT attention | Universal representations on Meta-Dataset · VGGFlower | accuracy: 82.8$\pm$0.7 -> 88.2$\pm$0.6 | Hurt |
| Add support-set-conditioned learned keys (ablation) | URT · Birds | accuracy: -18.1 | Helped |
| Add support-set-conditioned learned keys (ablation) | URT · Fungi | accuracy: -22.4 | Helped |
| Add a learned query transformation W^q (ablation) | URT · Birds | accuracy: -0.1 | No clear effect |
| Add the attention-score regularizer Ω(Θ) (ablation) | URT · VGGFlower | accuracy: -9.0 | Helped |
| Use 2 attention heads instead of 1 (ablation) | URT · Meta-Dataset validation | average accuracy: 74.605 -> 77.145 | Helped |
| Increase the attention-head count from 2 to 8 (ablation) | URT · Meta-Dataset validation | average accuracy: 77.145 -> 74.473 | Hurt |
| Use URT rather than SUR with FiLM-modulated parametric-family backbones | SUR-pf · TrafficSigns | accuracy: 65.1 $\pm$ 0.8 -> 67.5 $\pm$ 0.8 | Helped |
Closest papers by meaning
- Cross-Domain Few-Shot Classification via Learned Feature-Wise Transformation (ICLR 2020 · similarity 0.89)
- Unsupervised Meta-Learning for Few-Shot Image Classification (NeurIPS 2019 · similarity 0.88)
- Pushing the Limits of Simple Pipelines for Few-Shot Learning: External Data and Fine-Tuning Make a Difference (CVPR 2022 · similarity 0.88)
- Pushing the Limit of Fine-Tuning for Few-Shot Learning: Where Feature Reusing Meets Cross-Scale Attention (AAAI 2024 · similarity 0.88)
- One Meta-tuned Transformer is What You Need for Few-shot Learning (ICML 2024 · similarity 0.88)
- Single Domain Generalization for Few-Shot Counting via Universal Representation Matching (CVPR 2025 · similarity 0.87)
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.