Research map / Label-efficient and robust learning / Recognition and long-tail
Learning Compositional Representations for Few-Shot Recognition
Pavel Tokmakov, Yu-Xiong Wang, Martial Hebert
Read the paperCheck your idea against 137,217 papers
AI-extracted. The measured effects below were read by AI from the paper's full text (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 compositionality regularization (soft, orthogonal) (ablation) | cosine classifier · CUB novel 1-shot | top-5 accuracy: 46.1 -> 52.5 | Improved |
| add compositionality regularization (soft, orthogonal) (ablation) | cosine classifier · SUN novel 1-shot | top-5 accuracy: 35.4 -> 43.4 | Improved |
| add compositionality regularization on all classes (ablation) | cosine classifier · CUB novel+base 1-shot | top-5 accuracy: 58.2 -> 62.6 | Improved |
| use cosine instead of linear classifier (ablation) | compositional representation · CUB novel 1-shot | top-5 accuracy: 47.0 -> 52.5 | Improved |
| add data augmentation in few-shot training (ablation) | Cos w/ comp · CUB novel 1-shot | top-5 accuracy: 52.5 -> 53.6 | Improved |
| add data augmentation in few-shot training (ablation) | Cos baseline · SUN novel 1-shot | top-5 accuracy: 35.4 -> 39.9 | Improved |
| use hard sum constraint instead of none (ablation) | baseline · SUN novel | top-5 accuracy | Hurt |
| use soft constraint instead of hard constraint (ablation) | compositional regularization · CUB and SUN novel | top-5 accuracy | Improved |
| add orthogonality constraint on attribute embeddings (ablation) | soft compositional regularization · CUB and SUN novel | top-5 accuracy | Improved |
| reduce attributes to 25% on CUB and 50% on SUN (ablation) | compositional regularization · CUB and SUN novel | top-5 accuracy | No clear effect |
| add compositional representation to Matching Networks (ablation) | Matching Networks · CUB novel 1-shot | top-5 accuracy: 48.5 -> 50.9 | Improved |
| add compositional representation to Prototypical Networks (ablation) | Prototypical Networks · CUB novel 1-shot | top-5 accuracy: 43.2 -> 41.5 | Hurt |
Closest papers by meaning
- Does Data Scaling Lead to Visual Compositional Generalization? (ICML 2025 · similarity 0.89)
- When does compositional structure yield compositional generalization? A kernel theory. (ICLR 2025 · similarity 0.88)
- Provable Compositional Generalization for Object-Centric Learning (ICLR 2024 · similarity 0.87)
- Compositional Few-Shot Class-Incremental Learning (ICML 2024 · similarity 0.87)
- Towards Compositionality in Concept Learning (ICML 2024 · similarity 0.87)
- Not Only Text: Exploring Compositionality of Visual Representations in Vision-Language Models (CVPR 2025 · 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.
