Research map / Architectures and efficiency / Pruning and efficient tuning
AdaFilter: Adaptive Filter Fine-Tuning for Deep Transfer 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 per-example adaptive filter selection instead of standard fine-tuning | ResNet-50 transfer learning · Seven image-classification datasets | accuracy: 77.47% -> 82.44%; 73.10% -> 76.99%; 52.59% -> 55.41%; 78.09% -> 80.62%; 82.25% -> 84.31%; 76.42% -> 77.53%; 87.06% -> 87 | Helped |
| Replace Gated BN with Standard BN (ablation) | AdaFilter · Seven image-classification datasets | accuracy: 82.44% -> 82.02%; 55.41% -> 54.33%; 87.46% -> 87.27%; 76.99% -> 76.02%; 77.53% -> 77.01%; 80.06% -> 79.84%; 84.31% -> 83 | Hurt |
| Replace the recurrent gate with a CNN-based policy network (ablation) | AdaFilter · Image-classification datasets | accuracy: 82.44% -> 83.05%; 55.41% -> 54.63%; 87.46% -> 87.04%; 76.99% -> 76.33%; 77.53% -> 77.46%; 80.06% -> 80.25%; 84.31% -> 83 | Mixed |
| Use recurrent layerwise policy prediction rather than predicting all filter policies at once (ablation) | AdaFilter · Image-classification datasets | trainable parameter count | Helped |
| Use AdaFilter rather than standard fine-tuning during early training | ResNet-50 transfer learning · Stanford Dogs, after one epoch | accuracy: 54.69% -> 71.96% | Helped |
Closest papers by meaning
- 5%>100%: Breaking Performance Shackles of Full Fine-Tuning on Visual Recognition Tasks (CVPR 2025 · similarity 0.86)
- Neural Fine-Tuning Search for Few-Shot Learning (ICLR 2024 · similarity 0.86)
- Catch-Up Mix: Catch-Up Class for Struggling Filters in CNN (AAAI 2024 · similarity 0.86)
- Adaptive Random Feature Regularization on Fine-tuning Deep Neural Networks (CVPR 2024 · similarity 0.86)
- Deep Edge Filter: Return of the Human-Crafted Layer in Deep Learning (NeurIPS 2025 · similarity 0.85)
- Skip Tuning: Pre-trained Vision-Language Models are Effective and Efficient Adapters Themselves (CVPR 2025 · 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.