Research map / Label-efficient and robust learning / Zero-/few-shot and meta-learning
A Theoretical Analysis of the Number of Shots in Few-Shot Learning
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 |
|---|---|---|---|
| Apply EST to ProtoNet embeddings | ProtoNet, miniImageNet-5-way, 4-layer CNN, training shots 5 · miniImageNet-5-way, 1-10-shot average | accuracy: 65.08 ± 0.23 % -> 66.60 ± 0.23 % | Helped |
| Apply EST instead of PCA projection | ProtoNet, miniImageNet-5-way, 4-layer CNN, training shots 5 · miniImageNet-5-way, 1-10-shot average | accuracy: 66.12 ± 0.24 % -> 66.60 ± 0.23 % | Helped |
| Apply EST to ProtoNet embeddings | ProtoNet, tieredImageNet-5-way, 4-layer CNN, training shots 5 · tieredImageNet-5-way, 1-10-shot average | accuracy: 64.05 ± 0.33 % -> 65.46 ± 0.32 % | Helped |
| Apply EST instead of PCA projection | ProtoNet, tieredImageNet-5-way, 4-layer CNN, training shots 5 · tieredImageNet-5-way, 1-10-shot average | accuracy: 64.96 ± 0.31 % -> 65.46 ± 0.32 % | Helped |
| Apply EST to ProtoNet embeddings | ProtoNet, miniImageNet-5-way, 7-layer ResNet, training shots 5 · miniImageNet-5-way, 1-10-shot average | accuracy: 67.18 ± 0.23 % -> 68.19 ± 0.23 % | Helped |
| Apply EST to ProtoNet embeddings | ProtoNet, tieredImageNet-5-way, 7-layer ResNet, training shots 5 · tieredImageNet-5-way, 1-10-shot average | accuracy: 66.99 ± 0.31 % -> 67.91 ± 0.31 % | Helped |
| Apply EST to ProtoNet embeddings | ProtoNet, Omniglot-20-way, 4-layer CNN, training shots 1 · Omniglot-20-way, 1-5-shot average | accuracy: 97.81 ± 0.06 % -> 97.83 ± 0.06 % | No clear effect |
| Apply EST to ProtoNet embeddings | ProtoNet, Omniglot-20-way, 7-layer ResNet, training shots 1 · Omniglot-20-way, 1-5-shot average | accuracy: 98.35 ± 0.05 % -> 98.19 ± 0.05 % | Hurt |
| Train using mixed shots 1-10 instead of fixed 5 shots | ProtoNet, miniImageNet-5-way, 4-layer CNN · miniImageNet-5-way, 1-10-shot average | accuracy: 65.08 ± 0.23 % -> 65.83 ± 0.21 % | Mixed |
| Train using 10 shots instead of 5 shots | Vanilla ProtoNet, miniImageNet-5-way, 4-layer CNN · miniImageNet-5-way, 1-10-shot average | accuracy: 65.08 ± 0.23 % -> 63.54 ± 0.25 % | Hurt |
| Add a fully connected layer producing 60-dimensional embeddings (ablation) | ProtoNet, miniImageNet-5-way, 4-layer CNN · miniImageNet-5-way, 1-10-shot average | accuracy: 65.08 ± 0.23 % -> 61.05 ± 0.28 % | Hurt |
| Use PCA projection instead of EST | ProtoNet, miniImageNet-5-way, 7-layer ResNet, training shots 5 · miniImageNet-5-way, 1-10-shot average | accuracy: 67.63 ± 0.23 % -> 68.19 ± 0.23 % | Hurt |
Closest papers by meaning
- Meta-Learning for Semi-Supervised Few-Shot Classification (ICLR 2018 · similarity 0.91)
- Meta-Dataset: A Dataset of Datasets for Learning to Learn from Few Examples (ICLR 2020 · similarity 0.91)
- Unraveling Meta-Learning: Understanding Feature Representations for Few-Shot Tasks (ICML 2020 · similarity 0.91)
- The Role of Global Labels in Few-Shot Classification and How to Infer Them (NeurIPS 2021 · 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)
- MetaCoCo: A New Few-Shot Classification Benchmark with Spurious Correlation (ICLR 2024 · 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.