Research map / Architectures and efficiency / Pruning and efficient tuning
TR-PTS: Task-Relevant Parameter and Token Selection for Efficient Tuning
Siqi Luo, Haoran Yang, Yi Xin, Mingyang Yi, Guangyang Wu, Guangtao Zhai, Xiaohong Liu
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 parameter selection (FIM, layer-wise) to linear baseline (ablation) | Linear probing on ViT-B/16 · VTAB-1k dSprites/loc | accuracy: 12.5 -> 85.1 | Improved |
| add token selection and merging to linear baseline (ablation) | Linear probing on ViT-B/16 · VTAB-1k dSprites/loc | accuracy: 12.5 -> 14.8 | Improved |
| add token selection on top of parameter selection (ablation) | TR-PTS parameter selection · VTAB-1k dSprites/loc | accuracy: 85.1 -> 87.7 | Improved |
| add token selection on top of parameter selection (ablation) | TR-PTS parameter selection · VTAB-1k Flower102 | accuracy: 99.4 -> 99.5 | Improved |
| add token selection on top of parameter selection (ablation) | TR-PTS parameter selection · VTAB-1k Sun397 | accuracy: 54.2 -> 54.5 | Improved |
| add parameter selection on top of token selection (ablation) | TR-PTS token selection · VTAB-1k Sun397 | accuracy: 51.2 -> 54.5 | Improved |
| place token selection in sparse-parameter layers instead of dense layers (ratio 0.95) (ablation) | TR-PTS · VTAB-1k dSprites/loc | accuracy: 85.2 -> 87.7 | Improved |
| place token selection in sparse-parameter layers instead of random layers (ratio 0.95) (ablation) | TR-PTS · VTAB-1k Patch Camelyon | accuracy: 87.9 -> 88.1 | Improved |
| lower token selection ratio from 0.95 to 0.8 in sparse layers (ablation) | TR-PTS · VTAB-1k Sun397 | accuracy: 54.5 -> 52.9 | Hurt |
| lower token selection ratio from 0.95 to 0.8 in sparse layers (ablation) | TR-PTS · VTAB-1k dSprites/loc | accuracy: 87.7 -> 86.0 | Hurt |
| TR-PTS versus GPS (gradient-based parameter selection) | TR-PTS · VTAB-1k (19 tasks) | mean accuracy: 75.18 -> 75.92 | Improved |
| TR-PTS versus GPS | TR-PTS · FGVC (5 datasets) | mean accuracy: 91.78 -> 91.94 | 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 — 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
- TADFormer: Task-Adaptive Dynamic TransFormer for Efficient Multi-Task Learning (CVPR 2025 · similarity 0.88)
- Sensitivity-Aware Efficient Fine-Tuning via Compact Dynamic-Rank Adaptation (CVPR 2025 · similarity 0.87)
- Lessons and Insights from a Unifying Study of Parameter-Efficient Fine-Tuning (PEFT) in Visual Recognition (CVPR 2025 · similarity 0.87)
- Tuning Stable Rank Shrinkage: Aiming at the Overlooked Structural Risk in Fine-tuning (CVPR 2024 · similarity 0.87)
- TTE: Two Tokens Are Enough to Improve Parameter-Efficient Tuning (AAAI 2025 · similarity 0.87)
- Adaptive Parameter Selection for Tuning Vision-Language Models (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.
