Research map / Label-efficient and robust learning / Zero-/few-shot and meta-learning
Reinforced Attention for Few-Shot Learning and Beyond
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 RAP attention to MAML | MAML with ResNet-10 · miniImageNet 5-way 1-shot | accuracy: 54.69±0.89 -> 56.13±0.62 | Helped |
| Add RAP attention to ProtoNet | ProtoNet with ResNet-10 · miniImageNet 5-way 5-shot | accuracy: 72.64±0.64 -> 74.54±0.45 | Helped |
| Add RAP attention to LaplacianShot | LaplacianShot with DenseNet-121 · miniImageNet 5-way 1-shot | accuracy: 75.57±0.19 -> 75.58±0.20 | No clear effect |
| Add RAP attention to MAML | MAML with ResNet-18 · CUB-200-2011 5-way 5-shot | accuracy: 83.47±0.62 -> 86.07±0.43 | Helped |
| Add RAP attention to ProtoNet | ProtoNet with ResNet-18 · CUB-200-2011 5-way 5-shot | accuracy: 86.64±0.51 -> 89.23±0.31 | Helped |
| Add RAP attention to ResNet-50 | Image classification model · CIFAR100 | accuracy: +1.14 | Helped |
| Set α to 1e-4 instead of α=0 (ablation) | RAP-ProtoNet Conv-4 · CUB-200-2011 5-way 1-shot | accuracy: 53.83±0.66 -> 56.71±0.66 | Helped |
| Set α to 1e-2 instead of α=1e-4 (ablation) | RAP-ProtoNet ResNet-10 · CUB-200-2011 5-way 1-shot | accuracy: 75.17±0.63 -> 72.61±0.64 | Hurt |
| Set α to 1 instead of α=1e-4 (ablation) | RAP-ProtoNet ResNet-10 · CUB-200-2011 5-way 1-shot | accuracy: 75.17±0.63 -> 66.33±0.64 | Hurt |
| Increase recurrent steps from T=2 to T=5 (ablation) | RAP-ProtoNet Conv-4 · CUB-200-2011 5-way 1-shot | accuracy: 54.01±0.65 -> 56.71±0.66 | Helped |
| Replace RAP policy attention with SENet (ablation) | ProtoNet Conv-4 · CUB-200-2011 5-way 1-shot | accuracy: 55.18±0.61 vs 56.71±0.66 | No clear effect |
| Replace RAP policy attention with CBAM (ablation) | ProtoNet ResNet-10 · CUB-200-2011 5-way 5-shot | accuracy: 86.70±0.48 vs 88.29±0.34 | Hurt |
Closest papers by meaning
- CAD: Co-Adapting Discriminative Features for Improved Few-Shot Classification (CVPR 2022 · similarity 0.88)
- Few-Shot Learning via Embedding Adaptation With Set-to-Set Functions (CVPR 2020 · similarity 0.88)
- Few-Shot Classification With Feature Map Reconstruction Networks (CVPR 2021 · similarity 0.87)
- A Dual Attention Network with Semantic Embedding for Few-Shot Learning (AAAI 2019 · similarity 0.87)
- Pushing the Limit of Fine-Tuning for Few-Shot Learning: Where Feature Reusing Meets Cross-Scale Attention (AAAI 2024 · similarity 0.86)
- Scaling Few-Shot Learning for the Open World (AAAI 2024 · 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.