Research map / Architectures and efficiency / Pruning and efficient tuning
SCT: A Simple Baseline for Parameter-Efficient Fine-Tuning via Salient Channels
Hengyuan Zhao, Pichao Wang, Yuyang Zhao, Hao Luo, Fan Wang, Mike Zheng Shou
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 class-aware importance score (ablation) | SCT · VTAB-1K (19 tasks avg) | average top-1 accuracy: 72.8 -> 73.6 | Improved |
| select salient (highest L2) channels instead of inconspicuous ones (ablation) | SCT · VTAB-1K (19 tasks avg) | average top-1 accuracy: 71.1 -> 73.6 | Improved |
| select salient channels instead of random ones (ablation) | SCT · VTAB-1K (19 tasks avg) | average top-1 accuracy: 72.0 -> 73.6 | Improved |
| insert SCTM after Attn instead of after MLP (ablation) | SCT · VTAB-1K (19 tasks avg) | average top-1 accuracy: 72.9 -> 73.6 | Improved |
| insert SCTM after both Attn and MLP (ablation) | SCT · VTAB-1K Cifar100 / Resisc45 / DMLab | top-1 accuracy: 75.3 -> 75.7 | Mixed |
| increase K from 96 to 192 (ablation) | SCT · VTAB-1K (19 tasks avg) | average top-1 accuracy: 73.6 -> 74.0 | Improved |
| increase K from 32 to 96 (ablation) | SCT · VTAB-1K (19 tasks avg) | average top-1 accuracy: 69.1 -> 73.6 | Improved |
| use K=32 instead of full fine-tuning | SCT · VTAB-1K (19 tasks avg) | average top-1 accuracy: 65.6 -> 69.1 | Improved |
| raise channels in first four layers from 96 to 192 (ablation) | SCT · VTAB-1K Caltech101 / Resisc45 / KITTI | top-1 accuracy: 91.6 -> 91.7 | Mixed |
| reduce K from 96 to 32 in last eight layers (ablation) | SCT · VTAB-1K KITTI | top-1 accuracy: 80.2 -> 76.2 | Hurt |
| insert SCTM into more of the last layers (ablation) | SCT · VTAB-1K Resisc45 / DMLab | top-1 accuracy | Improved |
| use SCT instead of Adapter-8 | ViT-B/16 · VTAB-1K (19 tasks avg) | average top-1 accuracy: 71.3 -> 73.6 | Improved |
Earlier methods this paper compares with
Named baselines from earlier accepted papers in this paper's results tables: better than 1 of 1 on most shared results.
- AdaptFormer-8 — AdaptFormer: Adapting Vision Transformers for Scalable Visual Recognition (NeurIPS 2022) · better on 2 of 2 shared results
Matched by method name in results tables read by AI. Authors' own tables tend to favour the newer method, and a comparison can be missed when a method is renamed.
Closest papers by meaning
- Facing the Elephant in the Room: Visual Prompt Tuning or Full finetuning? (ICLR 2024 · similarity 0.88)
- Strong Baselines for Parameter-Efficient Few-Shot Fine-Tuning (AAAI 2024 · similarity 0.87)
- DA-VPT: Semantic-Guided Visual Prompt Tuning for Vision Transformers (CVPR 2025 · similarity 0.87)
- WST: Wavelet-Based Multi-scale Tuning for Visual Transfer Learning (AAAI 2025 · similarity 0.87)
- On the Surprising Effectiveness of Attention Transfer for Vision Transformers (NeurIPS 2024 · similarity 0.86)
- Sensitivity-Aware Efficient Fine-Tuning via Compact Dynamic-Rank Adaptation (CVPR 2025 · similarity 0.86)
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.
