Research map / Label-efficient and robust learning / Zero-/few-shot and meta-learning
Partial Is Better Than All: Revisiting Fine-tuning Strategy for Few-shot Learning
Read the paperCheck 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 |
|---|---|---|---|
| Use searched partial fine-tuning instead of freezing the backbone (ablation) | Baseline++ with ResNet-12 · mini-ImageNet 1-shot | accuracy: 59.35±0.82 -> 64.21±0.77 | Helped |
| Use searched partial fine-tuning instead of freezing the backbone (ablation) | Baseline++ with ResNet-12 · mini-ImageNet 5-shot | accuracy: 77.51±0.59 -> 80.38±0.59 | Helped |
| Use searched partial fine-tuning instead of freezing the backbone (ablation) | ProtoNet with ResNet-12 · CUB | accuracy: 1-shot: 73.82±0.92 -> 73.16±0.92; 5-shot: 87.28±0.48 -> 88.32±0.46 | Mixed |
| Use searched partial fine-tuning instead of freezing the backbone (ablation) | ProtoNet with ResNet-12 · mini-ImageNet | accuracy: 1-shot: 53.88±0.81 -> 54.36±0.81; 5-shot: 74.87±0.67 -> 76.59±0.64 | Helped |
| Fine-tune only the last convolutional layer instead of freezing the backbone (ablation) | Conv6 non-meta method · CUB 1-shot | accuracy: 66.75±0.91 -> 64.27±0.97 | Hurt |
| Search partial fine-tuning rather than use the fixed strategy (ablation) | Conv6 non-meta method · CUB | accuracy: 1-shot: 66.75±0.91 -> 65.82±0.97; 5-shot: 79.48±0.63 -> 80.48±0.60 | Mixed |
| Use GroupNorm instead of BatchNorm (ablation) | Baseline++ with searched Conv6 · mini-ImageNet to CUB, 5-shot | accuracy: 61.53±0.70 -> 67.42±0.72 | Helped |
| Use GroupNorm instead of BatchNorm (ablation) | ProtoNet with searched Conv6 · mini-ImageNet to CUB | accuracy: 1-shot: 36.36±0.73 -> 38.36±0.69; 5-shot: 54.06±0.73 -> 62.33±0.75 | Mixed |
| Partially reinitialize weights rather than inherit all weights (ablation) | Inception V3 transfer learning · ImageNet to CUB | Top-1 Accuracy: 82.9% -> 83.8% | Helped |
Closest papers by meaning
- Neural Fine-Tuning Search for Few-Shot Learning (ICLR 2024 · similarity 0.91)
- Pushing the Limit of Fine-Tuning for Few-Shot Learning: Where Feature Reusing Meets Cross-Scale Attention (AAAI 2024 · similarity 0.89)
- Strong Baselines for Parameter-Efficient Few-Shot Fine-Tuning (AAAI 2024 · similarity 0.89)
- A Baseline for Few-Shot Image Classification (ICLR 2020 · similarity 0.89)
- A Closer Look at Few-shot Classification Again (ICML 2023 · similarity 0.89)
- Simple Semantic-Aided Few-Shot Learning (CVPR 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.