Research map / Label-efficient and robust learning / Zero-/few-shot and meta-learning
Finding Task-Relevant Features for Few-Shot Learning by Category Traversal
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 CTM to the matching network | Matching Network · mini and tiered, 5-way and 20-way few-shot | accuracy: +3.54, +3.74, +2.66, +4.25; +2.99, +3.34, +2.23, +3.42 | Helped |
| Add CTM to the prototypical network | Prototypical Network · mini and tiered, 5-way and 20-way few-shot | accuracy: +3.23, +3.79, +3.55, +4.75; +3.50, +3.44, +2.46, +2.10 | Helped |
| Add CTM to the relation network | Relation Network · mini and tiered, 5-way and 20-way few-shot | accuracy: +3.84, +4.34, +3.76, +3.53; +3.67, +3.66, +4.76, +4.36 | Helped |
| Add the learned concentrator network (ablation) | CTM with ResNet-18 · mini, 1-shot | accuracy: 55.41 -> 59.34 | Helped |
| Add the learned concentrator network (ablation) | CTM with ResNet-18 · mini, 5-shot | accuracy: 73.29 -> 77.95 | Helped |
| Add the projector mask (ablation) | CTM with ResNet-18 · mini, 1-shot | accuracy: 57.18 -> 59.34 | Helped |
| Add the projector mask (ablation) | CTM with ResNet-18 · mini, 5-shot | accuracy: 74.25 -> 77.95 | Helped |
| Apply projector softmax across all locations rather than separately across channels at each location (ablation) | CTM with ResNet-18 · mini, 1-shot | accuracy: 59.34 -> 57.77 | Hurt |
| Apply projector softmax across all locations rather than separately across channels at each location (ablation) | CTM with ResNet-18 · mini, 5-shot | accuracy: 77.95 -> 75.03 | Hurt |
| Use sample-wise masking rather than cluster-wise masking (ablation) | CTM with 4-layer backbone · mini, 5-way 1-shot | accuracy: 39.55% -> 41.62% | Helped |
| Use sample-wise masking rather than cluster-wise masking (ablation) | CTM with 4-layer backbone · mini, 5-way 5-shot | accuracy: 56.95% -> 58.77% | Helped |
| Use cross-entropy rather than MSE loss (ablation) | Relation Network with CTM · mini, 1-shot | accuracy: 61.37 -> 62.05; 0.68% | Helped |
Closest papers by meaning
- DiffKendall: A Novel Approach for Few-Shot Learning with Differentiable Kendall's Rank Correlation (NeurIPS 2023 · similarity 0.90)
- TADAM: Task dependent adaptive metric for improved few-shot learning (NeurIPS 2018 · similarity 0.90)
- Pushing the Limit of Fine-Tuning for Few-Shot Learning: Where Feature Reusing Meets Cross-Scale Attention (AAAI 2024 · similarity 0.88)
- Rethinking Few-Shot Adaptation of Vision-Language Models in Two Stages (CVPR 2025 · similarity 0.88)
- Few-Shot Recognition via Stage-Wise Retrieval-Augmented Finetuning (CVPR 2025 · similarity 0.88)
- FeatWalk: Enhancing Few-Shot Classification through Local View Leveraging (AAAI 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.