Research map / Label-efficient and robust learning / Zero-/few-shot and meta-learning
Boosting Few-Shot Learning With Adaptive Margin Loss
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 TRAML to AM3 | AM3 (Prototypical Network) · miniImageNet 5-way 1-shot | accuracy: 65.21±0.49 -> 67.10 ± 0.52 | Helped |
| Add TRAML to AM3 | AM3 (Prototypical Network) · miniImageNet 5-way 5-shot | accuracy: 75.20 ± 0.36 -> 79.54 ± 0.60 | Helped |
| Add TRAML to Dynamic FSL | Dynamic FSL · ImageNet2012 generalized FSL, novel classes, 1-, 2-, 5-, 10-, and 20-shot | top-5 accuracy: 46.0, 57.5, 69.2, 74.8, 78.1 -> 48.1, 59.2, 70.3, 76.4, 79.4 | Helped |
| Replace original classification loss with a fixed additive margin (ablation) | AM3 (Prototypical Network) · miniImageNet 5-way 1-shot | accuracy: 65.21 ± 0.49 -> 65.42 ± 0.25 | Helped |
| Replace original classification loss with a fixed additive margin (ablation) | AM3 (Prototypical Network) · miniImageNet 5-way 5-shot | accuracy: 75.20 ± 0.36 -> 75.48 ± 0.34 | Helped |
| Use class-pair semantic similarity to adapt margins (ablation) | AM3 with naive additive margin loss · miniImageNet 5-way 1-shot | accuracy: 65.42 ± 0.25 -> 66.36 ± 0.57 | Helped |
| Use class-pair semantic similarity to adapt margins (ablation) | AM3 with naive additive margin loss · miniImageNet 5-way 5-shot | accuracy: 75.48 ± 0.34 -> 77.21 ± 0.48 | Helped |
| Add task-level semantic context to class-relevant margins (ablation) | AM3 with class-relevant additive margin loss · miniImageNet 5-way 1-shot | accuracy: 66.36 ± 0.57 -> 67.10 ± 0.52 | Helped |
| Add task-level semantic context to class-relevant margins (ablation) | AM3 with class-relevant additive margin loss · miniImageNet 5-way 5-shot | accuracy: 77.21 ± 0.48 -> 79.54 ± 0.60 | Helped |
| Use TRAML instead of additive angular margin loss | AM3 (Prototypical Network) · miniImageNet 5-way 1-shot | accuracy: 66.21±0.46 -> 67.10 ± 0.52 | Helped |
| Use TRAML instead of additive cosine margin loss | AM3 (Prototypical Network) · miniImageNet 5-way 5-shot | accuracy: 76.93 ± 0.49 -> 79.54 ± 0.60 | Helped |
| Use cosine distance instead of Euclidean distance (ablation) | AM3 with TRAML · miniImageNet 5-way 1-shot | accuracy: 67.10 ± 0.52 -> 66.92 ± 0.43 | Hurt |
Closest papers by meaning
- TADAM: Task dependent adaptive metric for improved few-shot learning (NeurIPS 2018 · similarity 0.89)
- Meta-Learning With Differentiable Convex Optimization (CVPR 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)
- Boosting Few-Shot Learning via Attentive Feature Regularization (AAAI 2024 · similarity 0.88)
- Discriminative Sample-Guided and Parameter-Efficient Feature Space Adaptation for Cross-Domain Few-Shot Learning (CVPR 2024 · similarity 0.87)
- Few-Shot Recognition via Stage-Wise Retrieval-Augmented Finetuning (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.