Research map / Label-efficient and robust learning / Zero-/few-shot and meta-learning
Dynamic Few-Shot Visual Learning Without Forgetting
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 |
|---|---|---|---|
| use the attention-based weight generator with the cosine classifier rather than feature averaging alone (ablation) | Cosine Classifier · Mini-ImageNet validation, 1-shot, K_novel=5 | novel accuracy: 55.33 ± 0.46% -> 58.55 ± 0.50% | Helped |
| replace the dot-product classifier with a cosine-similarity classifier (ablation) | ConvNet recognition model · Mini-ImageNet validation, 5-shot, K_novel=5 | novel accuracy: 64.58 ± 0.38% -> 72.83 ± 0.35% | Helped |
| add the attention-based few-shot weight generator (ablation) | Cosine Classifier · Mini-ImageNet validation, 5-shot, K_novel=5 | novel accuracy: 72.83 ± 0.35% -> 74.92 ± 0.36% | Helped |
| add the attention-based few-shot weight generator (ablation) | Cosine Classifier · Mini-ImageNet validation, 1-shot, K_novel=5 | novel accuracy: 54.55 ± 0.44% -> 58.55 ± 0.50% | Helped |
| remove the final ReLU from the feature extractor (ablation) | Cosine-similarity ConvNet · Mini-ImageNet validation, 5-shot, K_novel=5 | novel accuracy: 71.04 ± 0.36% -> 74.92 ± 0.36% | Helped |
| remove the final ReLU from the feature extractor (ablation) | Cosine-similarity ConvNet · Mini-ImageNet validation, base categories | base accuracy: 72.51% -> 70.88% | Hurt |
Closest papers by meaning
- Few-Shot Image Recognition by Predicting Parameters From Activations (CVPR 2018 · similarity 0.92)
- A Closer Look at Few-shot Classification Again (ICML 2023 · similarity 0.90)
- Adaptive Subspaces for Few-Shot Learning (CVPR 2020 · similarity 0.90)
- Generalized Few-Shot Object Detection Without Forgetting (CVPR 2021 · similarity 0.89)
- Few-Shot Learning via Embedding Adaptation With Set-to-Set Functions (CVPR 2020 · similarity 0.89)
- Pushing the Limit of Fine-Tuning for Few-Shot Learning: Where Feature Reusing Meets Cross-Scale Attention (AAAI 2024 · similarity 0.89)
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.