Research map / Architectures and efficiency / Pruning and efficient tuning
Efficient Adaptation of Pre-trained Vision Transformer underpinned by Approximately Orthogonal Fine-Tuning Strategy
Yiting Yang, Hao Luo, Yuan Sun, Qingsen Yan, Haokui Zhang, Wei Dong, Guoqing Wang, Peng Wang, Yang Yang, Hengtao Shen
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 |
|---|---|---|---|
| replace LoRA matrices with AOFT | LoRA · VTAB-1k ViT-B/16 | mean accuracy: 72.3 -> 72.9 | Improved |
| replace Adapter matrices with AOFT | Adapter · VTAB-1k ViT-B/16 | mean accuracy: 71.4 -> 72.5 | Improved |
| replace Adapter matrices with AOFT | Adapter · FGVC ViT-B/16 | mean accuracy: 85.7 -> 88.9 | Improved |
| replace VPT-Shallow prompts with AOFT | VPT-Shallow · FGVC ViT-B/16 | mean accuracy: 84.7 -> 84.0 | Hurt |
| replace VPT-Shallow prompts with AOFT | VPT-Shallow · VTAB-1k ViT-B/16 | mean accuracy: 64.8 -> 66.8 | Improved |
| replace VPT-Deep prompts with AOFT | VPT-Deep · VTAB-1k ViT-B/16 | mean accuracy: 69.4 -> 70.7 | Improved |
| replace LoRA with AOFT* on ViT-L | LoRA · VTAB-1k ViT-L | mean accuracy: 72.0 -> 74.3 | Improved |
| replace LoRA with AOFT* on ViT-H | LoRA · VTAB-1k ViT-H | mean accuracy: 69.3 -> 71.3 | Improved |
| replace Adapter with AOFT* on ViT-L | Adapter · VTAB-1k ViT-L | mean accuracy: 52.9 -> 60.9 | Improved |
| add learnable scaling vector (AOFT*) to LoRA+AOFT (ablation) | LoRA+AOFT · VTAB-1k ViT-B/16 | mean accuracy: 72.9 -> 71.5 | Hurt |
| add learnable scaling vector (AOFT*) to Adapter+AOFT (ablation) | Adapter+AOFT · FGVC ViT-B/16 | mean accuracy: 88.9 -> 90.1 | Improved |
| apply AOFT also to W_FC1 and W_FC2 in LoRA (ablation) | LoRA+AOFT (Wq,Wv) · VTAB-1k ViT-B/16 | mean accuracy: 73.1 -> 73.4 | Improved |
Earlier methods this paper compares with
Named baselines from earlier accepted papers in this paper's results tables: better than 2 of 2 on most shared results.
- VeRA — VeRA: Vector-based Random Matrix Adaptation (ICLR 2024) · better on 1 of 1 shared result
- DoRA — DoRA: Weight-Decomposed Low-Rank Adaptation (ICML 2024) · better on 1 of 1 shared result
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
- Efficient Adaptation of Pre-trained Vision Transformer via Householder Transformation (NeurIPS 2024 · similarity 0.93)
- Low-Rank Rescaled Vision Transformer Fine-Tuning: A Residual Design Approach (CVPR 2024 · similarity 0.90)
- Structured Unrestricted-Rank Matrices for Parameter Efficient Finetuning (NeurIPS 2024 · similarity 0.88)
- RandLoRA: Full rank parameter-efficient fine-tuning of large models (ICLR 2025 · similarity 0.88)
- PELA: Learning Parameter-Efficient Models with Low-Rank Approximation (CVPR 2024 · similarity 0.88)
- Sensitivity-Aware Efficient Fine-Tuning via Compact Dynamic-Rank Adaptation (CVPR 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.
