Research map / Label-efficient and robust learning / Zero-/few-shot and meta-learning
The Role of Global Labels in Few-Shot Classification and How to Infer Them
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 |
|---|---|---|---|
| Infer global labels using local-task constraints, then meta-learn the embedding (ablation) | MeLa · miniImageNet, tieredImageNet | accuracy: 60.2 \pm 0.3 -> 62.0 \pm 0.4; 75.6 \pm 0.5 -> 79.6 \pm 0.3; 64.3\pm 0.5 -> 69.1 \pm 0.5; 78.9 \pm 0.4 -> 84.2 \pm 0.3 | Helped |
| Replace local-constraint clustering with unconstrained K-means (ablation) | MeLa · miniImageNet and tieredImageNet | accuracy: 62.0\pm 0.4 -> 60.7\pm 0.5; 79.6\pm 0.3 -> 76.9\pm 0.3; 69.0\pm 0.5 -> 64.8\pm 0.6; 84.1\pm 0.3 -> 78.8\pm 0.5 | Hurt |
| Train with examples sampled without replacement (ablation) | MeLa · miniImageNet and tieredImageNet | accuracy: 62.0 \pm 0.4 -> 61.8 \pm 0.5; 79.6 \pm 0.3 -> 79.4 \pm 0.4; 69.1 \pm 0.5 -> 69.1 \pm 0.5; 84.2 \pm 0.3 -> 84.1 \pm 0.3 | No clear effect |
| Add S2M2 augmentation and fine-tuning after inferred labels | MeLa · miniImageNet and tieredImageNet | accuracy: +2.8 accuracy points at miniImageNet 1-shot versus MeLa; +3.5 at 5-shot; +1.5 at tieredImageNet 1-shot; +1.7 at 5-shot | Helped |
| Vary the pruning threshold q (ablation) | MeLa labeler · miniImageNet and tieredImageNet | accuracy: miniImageNet: 60.9\pm 0.5, 60.9\pm 0.5, 62.0 \pm 0.4; tieredImageNet: 69.1 \pm 0.5, 68.5\pm 0.4, 68.4\pm 0.4 | No clear effect |
| Use inferred global labels instead of inaccessible ground-truth global labels | MeLa versus RFS · miniImageNet and tieredImageNet | accuracy: miniImageNet: 62.0 \pm 0.4 vs 62.0 \pm 0.4, 79.6 \pm 0.3 vs 79.6 \pm 0.3; tieredImageNet: 69.1 \pm 0.5 vs 69.4 \pm 0.5, | Mixed |
Closest papers by meaning
- Unraveling Meta-Learning: Understanding Feature Representations for Few-Shot Tasks (ICML 2020 · similarity 0.91)
- Meta-Dataset: A Dataset of Datasets for Learning to Learn from Few Examples (ICLR 2020 · similarity 0.91)
- A Theoretical Analysis of the Number of Shots in Few-Shot Learning (ICLR 2020 · similarity 0.91)
- Pushing the Limit of Fine-Tuning for Few-Shot Learning: Where Feature Reusing Meets Cross-Scale Attention (AAAI 2024 · similarity 0.89)
- Making Large Vision Language Models to Be Good Few-Shot Learners (AAAI 2025 · similarity 0.88)
- MetaCoCo: A New Few-Shot Classification Benchmark with Spurious Correlation (ICLR 2024 · similarity 0.88)
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.