Research map / Label-efficient and robust learning / Zero-/few-shot and meta-learning
Meta-Transfer Learning for Few-Shot 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 HT hard-task resampling to MTL (ablation) | MTL with ResNet-12 (pre) · miniImageNet 5-way 1-shot | accuracy: 60.2 ± 1.8 -> 61.2 ± 1.8 | Helped |
| Add HT hard-task resampling to MTL (ablation) | MTL with ResNet-12 (pre) · FC100 5-way 1-shot | accuracy: 43.6 ± 1.8 -> 45.1 ± 1.8 | Helped |
| Add HT hard-task resampling to MTL (ablation) | MTL with ResNet-12 (pre) · FC100 5-way 5-shot | accuracy: 55.4 ± 0.9 -> 57.6 ± 0.9 | Helped |
| Add HT hard-task resampling to MTL (ablation) | MTL with ResNet-12 (pre) · FC100 5-way 10-shot | accuracy: 62.4 ± 0.8 -> 63.4 ± 0.8 | Helped |
| Use scaling and shifting instead of fine-tuning all feature extractor and classifier parameters (ablation) | MTL with ResNet-12 (pre) · miniImageNet 5-way 1-shot | accuracy: 58.3 -> 60.2 | Helped |
| Use scaling and shifting instead of fine-tuning all feature extractor and classifier parameters (ablation) | MTL with ResNet-12 (pre) · FC100 5-way 1-shot | accuracy: 41.6 -> 43.6 | Helped |
| Apply scaling and shifting to all feature extractor layers rather than only the fourth residual block (ablation) | MTL with ResNet-12 (pre) · miniImageNet 5-way 1-shot | accuracy: 59.2 -> 60.2 | Helped |
| Use a single-layer classifier instead of a 2-layer classifier (ablation) | MTL with ResNet-12 (pre) · miniImageNet 5-way 5-shot | accuracy: 70.7 ± 0.9 -> 74.3 ± 0.9 | Helped |
| Use a single-layer classifier instead of a 3-layer classifier (ablation) | MTL with ResNet-12 (pre) · FC100 5-way 10-shot | accuracy: 53.8 ± 0.8 -> 62.4 ± 0.8 | Helped |
| Use a single-layer classifier instead of a 2-layer classifier (ablation) | MTL with ResNet-12 (pre) · FC100 5-way 10-shot | accuracy: 54.1 ± 0.8 -> 62.4 ± 0.8 | Helped |
| Freeze the feature extractor during base-learning instead of updating it with the classifier (ablation) | MTL with ResNet-12 (pre) · FC100 5-way 10-shot | accuracy: 60.7 ± 0.8 -> 62.4 ± 0.8 | Helped |
| Use pre-trained ResNet-12 and MTL rather than MAML with 4CONV and HT meta-batches | Few-shot learning on miniImageNet · miniImageNet 5-way 1-shot | accuracy: 49.1 ± 1.9 -> 61.2 ± 1.8 | Helped |
Closest papers by meaning
- Bayesian Meta-Learning for the Few-Shot Setting via Deep Kernels (NeurIPS 2020 · similarity 0.91)
- A Unified Few-Shot Classification Benchmark to Compare Transfer and Meta Learning Approaches (NeurIPS 2021 · similarity 0.90)
- Meta-Learning without Memorization (ICLR 2020 · similarity 0.89)
- A Dual Attention Network with Semantic Embedding for Few-Shot Learning (AAAI 2019 · similarity 0.89)
- LGM-Net: Learning to Generate Matching Networks for Few-Shot Learning (ICML 2019 · similarity 0.89)
- Pushing the Limit of Fine-Tuning for Few-Shot Learning: Where Feature Reusing Meets Cross-Scale Attention (AAAI 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.