Research map / Label-efficient and robust learning / Zero-/few-shot and meta-learning
Learning to Select Base Classes for Few-Shot Classification
Linjun Zhou, Peng Cui, Xu Jia, Shiqiang Yang, Qi Tian
Check 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 |
|---|---|---|---|
| Select base classes by optimizing Similarity Ratio with K=1 | Few-shot classification with pre-trained selection · ImageNet, Caltech256, and CUB-200-2011; 100-way, m=100 | Top-1 accuracy: 39.39% ± 0.82% -> 43.42% ± 0.78%; 49.55% ± 1.28% -> 55.41% ± 1.25%; 28.11% ± 0.44% -> 29.65% ± 0.82% | Improved |
| Use DomSim selection instead of random selection | Pre-trained selection · ImageNet, 100-way, m=100, 5-shot | Top-1 accuracy: 39.39% ± 0.82% -> 38.00% ± 0.36% | Hurt |
| Increase K from 1 to 3 in the cold-start Alg. 2 schedule (ablation) | Cold Start Selection · ImageNet, 100-way | Top-1 accuracy: 41.75% ± 0.59% -> 42.17% ± 0.67% | Improved |
| Increase K from 3 to 5 in the cold-start Alg. 2 schedule (ablation) | Cold Start Selection · ImageNet, 100-way | Top-1 accuracy: 42.17% ± 0.67% -> 41.33% ± 0.36% | Hurt |
| Use an incremental 6-12-25-50-100 selection schedule rather than 50-100 (ablation) | Cold Start Selection · ImageNet, 100-way | Top-1 accuracy: 40.88% ± 0.66% -> 42.17% ± 0.67% | Improved |
| Use 6-12-25-50-100 rather than 10-20-40-80-100 (ablation) | Cold Start Selection · ImageNet, 100-way | Top-1 accuracy: 41.61% ± 0.76% -> 42.17% ± 0.67% | Improved |
| Increase K from 3 to 10 in Alg. 2, λ=0 (ablation) | General Selection · Caltech256, 100-way, m=20, 20-shot | Top-1 accuracy: 40.72% ± 0.92% -> 41.18% ± 0.88% | Improved |
| Add a positive diversity term with λ=0.2 (ablation) | General Selection · CUB-200-2011, 100-way, m=100, 20-shot | Top-1 accuracy: 28.33% ± 0.57% -> 32.50% ± 0.58% | Improved |
| Increase the positive diversity weight from λ=0.2 to λ=0.5 (ablation) | General Selection · CUB-200-2011, 100-way, m=20, 20-shot | Top-1 accuracy: 16.61% ± 0.36% -> 17.09% ± 0.33% | Improved |
| Increase the positive diversity weight from λ=0.2 to λ=0.5 (ablation) | General Selection · CUB-200-2011, 100-way, m=100, 20-shot | Top-1 accuracy: 32.50% ± 0.58% -> 31.01% ± 0.58% | Hurt |
Closest papers by meaning
- A Closer Look at Few-shot Classification (ICLR 2019 · similarity 0.91)
- Impact of base dataset design on few-shot image classification (ECCV 2020 · similarity 0.91)
- Provably Improving Generalization of Few-shot models with Synthetic Data (ICML 2025 · similarity 0.89)
- A Closer Look at Few-shot Classification Again (ICML 2023 · similarity 0.89)
- Improved Few-Shot Visual Classification (CVPR 2020 · similarity 0.89)
- Few-Class Arena: A Benchmark for Efficient Selection of Vision Models and Dataset Difficulty Measurement (ICLR 2025 · similarity 0.88)
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.
