Research map / Label-efficient and robust learning / Zero-/few-shot and meta-learning
Alleviating the Sample Selection Bias in Few-shot Learning by Removing Projection to the Centroid
Jing Xu, Xu Luo, Xinglin Pan, Yanan Li, Wenjie Pei, Zenglin Xu
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 TCPR projection removal | Baseline++ with ResNet-18 · miniImageNet 5-way 1-shot | accuracy: 54.23 ± 0.46 -> 58.67 ± 0.41 | Improved |
| Add TCPR projection removal | S2M2 · tieredImageNet 5-way 1-shot | accuracy: 74.87 ± 0.43 -> 77.67 ± 0.41 | Improved |
| Add TCPR projection removal | S2M2 · miniImageNet 5-way 5-shot | accuracy: 83.50 ± 0.29 -> 84.51 ± 0.27 | Improved |
| Estimate the task centroid from support and query features rather than base-neighbor features (ablation) | TCPR · Meta-Dataset mini-test 5-way 1-shot | accuracy: 68.06 -> 69.52 | Improved |
| Estimate the task centroid from support features only rather than base-neighbor features (ablation) | TCPR · Meta-Dataset mini-test 5-way 1-shot | accuracy: 68.06 -> 66.88 | Hurt |
| Estimate the task centroid from support and query features rather than base-neighbor features (ablation) | TCPR · Meta-Dataset CUB 5-way 1-shot | accuracy: 51.87 -> 53.83 | Improved |
| Estimate the task centroid from support features only rather than base-neighbor features (ablation) | TCPR · Meta-Dataset CUB 5-way 1-shot | accuracy: 51.87 -> 50.48 | Hurt |
| Add TCPR projection removal | S2M2 · miniImageNet 2-way 1-shot | accuracy: 84.52 -> 88.39 | Improved |
| Add TCPR projection removal | S2M2 · miniImageNet 10-way 1-shot | accuracy: 50.68 -> 52.80 | Improved |
| Use more base neighbors to approximate the task centroid (ablation) | TCPR · Novel datasets with varying base-to-novel domain gaps | accuracy | Mixed |
Closest papers by meaning
- Improved Few-Shot Visual Classification (CVPR 2020 · similarity 0.87)
- Learning to Select Base Classes for Few-Shot Classification (CVPR 2020 · similarity 0.86)
- TAFSSL: Task-Adaptive Feature Sub-Space Learning for few-shot classification (ECCV 2020 · similarity 0.86)
- A Closer Look at Few-shot Classification Again (ICML 2023 · similarity 0.85)
- Discriminative Sample-Guided and Parameter-Efficient Feature Space Adaptation for Cross-Domain Few-Shot Learning (CVPR 2024 · similarity 0.85)
- Pushing the Limit of Fine-Tuning for Few-Shot Learning: Where Feature Reusing Meets Cross-Scale Attention (AAAI 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.
