Research map / Architectures and efficiency / Pruning and efficient tuning
Structured Hyperedge Adaptation for Parameter-Efficient Fine-Tuning of Vision Transformers
Edwin Kwadwo Tenagyei, Lei Wang, Ugochukwu Akpudo, Jun Zhou, Yongsheng Gao
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 hypergraph routing (hyperedge-space adaptation) to a token-wise adapter (ablation) | adapter · VTAB-1K (ViT-B/16) | average accuracy (%): 76.3 -> 77.6 | Improved |
| use HyperAdapter at matched parameter budget instead of enlarged token-wise adapter (ablation) | adapter · VTAB-1K (ViT-B/16) | average accuracy (%): 76.7 -> 77.6 | Improved |
| increase hyperedges from K=4 to K=8 (ablation) | HyperAdapter · VTAB-1K (ViT-B/16) | average accuracy (%): 77.2 -> 77.6 | Improved |
| increase hyperedges beyond K=8 (ablation) | HyperAdapter · VTAB-1K (ViT-B/16) | average accuracy (%): 77.6 -> 77.3 | Hurt |
| increase bottleneck r from 8 to 16 (ablation) | HyperAdapter · VTAB-1K (ViT-B/16) | average accuracy (%): 77.6 -> 77.2 | No clear effect |
| use parallel placement on Attn+MLP instead of pre placement (ablation) | HyperAdapter · VTAB-1K (ViT-B/16) | average accuracy (%): 77.1 -> 77.6 | Improved |
| adapt MLP branch as well as attention (parallel) (ablation) | HyperAdapter · VTAB-1K (ViT-B/16) | average accuracy (%): 76.8 -> 77.6 | Improved |
| aggregate hyperedges (MeanH) for CLS update instead of no update (ablation) | HyperAdapter · VTAB-1K (ViT-B/16) | average accuracy (%): 77.1 -> 77.6 | Improved |
| use HyperAdapter instead of RepAdapter on ViT-L | adapter · VTAB-1K (ViT-L/16) | average accuracy (%): 76.8 -> 77.7 | Improved |
| use HyperAdapter instead of RepAdapter on Swin-B | adapter · VTAB-1K (Swin-B) | average accuracy (%): 77.0 -> 77.6 | Improved |
| use HyperAdapter instead of RepAdapter for segmentation | adapter · ADE20K (ViT-L) | mIoU single-scale: 44.44 -> 45.20 | 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.
- Res-Tuning — Res-Tuning: A Flexible and Efficient Tuning Paradigm via Unbinding Tuner from Backbone (NeurIPS 2023) · better on 1 of 1 shared result
- AdaptFormer — AdaptFormer: Adapting Vision Transformers for Scalable Visual Recognition (NeurIPS 2022) · 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
- Dynamic Tuning Towards Parameter and Inference Efficiency for ViT Adaptation (NeurIPS 2024 · similarity 0.89)
- ETHER: Efficient Finetuning of Large-Scale Models with Hyperplane Reflections (ICML 2024 · similarity 0.86)
- Time- Memory- and Parameter-Efficient Visual Adaptation (CVPR 2024 · similarity 0.86)
- Vision Transformers with Self-Distilled Registers (NeurIPS 2025 · similarity 0.86)
- Efficient Adaptation of Pre-trained Vision Transformer via Householder Transformation (NeurIPS 2024 · similarity 0.86)
- TADFormer: Task-Adaptive Dynamic TransFormer for Efficient Multi-Task Learning (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.
