Research map / Architectures and efficiency / Pruning and efficient tuning
Growing a Brain: Fine-Tuning by Increasing Model Capacity
Check 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 |
|---|---|---|---|
| Increase model capacity by adding a layer or channels while fine-tuning | AlexNet · SUN-397 | accuracy: 55.93 -> 59.05 | Helped |
| Widen existing layers rather than only adding a new layer | AlexNet · SUN-397 | accuracy: 58.54 -> 58.95 | Helped |
| Add channels at both FC6 and FC7 (ablation) | AlexNet · SUN-397 | accuracy: 58.95 -> 59.05 | Helped |
| Add normalization and adaptive scaling to new FC7 units (ablation) | WA-CNN · SUN-397 | accuracy: 51.34 -> 58.95 | Helped |
| Add normalization and adaptive scaling to the new layer (ablation) | DA-CNN · SUN-397 | accuracy: 57.21 -> 58.07 | Helped |
| Progressively widen instead of fine-tuning a fixed wide model (ablation) | AlexNet · MIT-67 | accuracy: 62.3 -> 66.3 | Helped |
| Progressively widen instead of fine-tuning a fixed wide model (ablation) | AlexNet · CUB200-2011 | accuracy: 63.2 -> 69.0 | Helped |
| Progressively widen rather than train the wide network from scratch (ablation) | AlexNet · ImageNet | accuracy: 57.6 -> 57.8 | Helped |
| Transfer progressively through SUN-397 before adapting to MIT-67 | WA-CNN · MIT-67 | accuracy: 66.3 -> 79.3 | Helped |
Closest papers by meaning
- Rethinking the Hyperparameters for Fine-tuning (ICLR 2020 · similarity 0.89)
- Tuning Stable Rank Shrinkage: Aiming at the Overlooked Structural Risk in Fine-tuning (CVPR 2024 · similarity 0.88)
- PROFIT: A Specialized Optimizer for Deep Fine Tuning (NeurIPS 2025 · similarity 0.88)
- Pay Attention to Small Weights (NeurIPS 2025 · similarity 0.88)
- Neural Fine-Tuning Search for Few-Shot Learning (ICLR 2024 · similarity 0.87)
- LiNeS: Post-training Layer Scaling Prevents Forgetting and Enhances Model Merging (ICLR 2025 · 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.