Research map / Label-efficient and robust learning / Zero-/few-shot and meta-learning
Negative Margin Matters: Understanding Margin in Few-shot Classification
Bin Liu, Yue Cao, Yutong Lin, Qi Li, Zheng Zhang, Mingsheng Long, Han Hu
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 negative margin to softmax loss (ablation) | softmax pre-training · mini-ImageNet validation classes, 1-shot | accuracy: 45.98 -> 56.95 | Improved |
| add negative margin to cosine softmax loss (ablation) | cosine softmax pre-training · mini-ImageNet validation classes, 1-shot | accuracy: 59.49 -> 63.68 | Improved |
| add negative margin to cosine softmax loss (ablation) | cosine softmax pre-training · CUB validation classes, 1-shot | accuracy: 66.39 -> 69.17 | Improved |
| add negative margin to cosine softmax loss (ablation) | cosine softmax pre-training · mini-ImageNet->CUB validation, 5-shot | accuracy: 61.99 -> 64.04 | Improved |
| add negative margin to softmax loss (ablation) | softmax pre-training · CUB validation classes, 5-shot | accuracy: 80.21 -> 80.60 | No clear effect |
| add negative margin (ablation) | Neg-Cosine ResNet-18 · mini-ImageNet novel, 1-shot | accuracy: 54.51 -> 60.25 | Improved |
| add weight decay on top of negative margin (ablation) | Neg-Cosine ResNet-18 · mini-ImageNet novel, 1-shot | accuracy: 60.25 -> 62.21 | Improved |
| add DropBlock on top of negative margin and weight decay (ablation) | Neg-Cosine ResNet-18 · mini-ImageNet novel, 1-shot | accuracy: 62.21 -> 62.33 | No clear effect |
| add label smoothing (ablation) | baseline softmax · mini-ImageNet validation classes, 1-shot | accuracy: 59.49 -> 58.02 | Hurt |
| add label smoothing (ablation) | baseline softmax · mini-ImageNet base classes, 1-shot | accuracy: 85.45 -> 88.93 | Improved |
| add negative margin (ablation) | baseline softmax · mini-ImageNet validation classes, 5-shot | accuracy: 78.90 -> 81.60 | Improved |
| add negative margin (ablation) | baseline softmax · mini-ImageNet base classes, 1-shot | accuracy: 85.45 -> 79.92 | Hurt |
Closest papers by meaning
- Boosting Few-Shot Learning With Adaptive Margin Loss (CVPR 2020 · similarity 0.90)
- Restoring Negative Information in Few-Shot Object Detection (NeurIPS 2020 · similarity 0.88)
- Learning Towards The Largest Margins (ICLR 2022 · similarity 0.88)
- Rethinking Preventing Class-Collapsing in Metric Learning With Margin-Based Losses (ICCV 2021 · similarity 0.87)
- Beyond Max-Margin: Class Margin Equilibrium for Few-Shot Object Detection (CVPR 2021 · similarity 0.87)
- Discriminative Sample-Guided and Parameter-Efficient Feature Space Adaptation for Cross-Domain Few-Shot Learning (CVPR 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.
