Research map / Architectures and efficiency / Pruning and efficient tuning
Res-Tuning: A Flexible and Efficient Tuning Paradigm via Unbinding Tuner from Backbone
Read the paperCheck 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 |
|---|---|---|---|
| connect prompt tuner to FFN instead of the default adapter-on-FFN (ablation) | Single-Res-Tuner · CIFAR-100 | accuracy: 92.34 -> 92.68 | Helped |
| use two tuners (MHA and FFN adapters) instead of one (ablation) | Res-Tuning · CIFAR-100 | accuracy: 92.68 -> 93.25 | Helped |
| add block-level tuner on top of Dual-Res-Tuner (ablation) | Res-Tuning · CIFAR-100 | accuracy: 93.25 -> 93.28 | Helped |
| reformulate PETL methods in unbinding form (ablation) | Adapter/Prefix/Prompt · CIFAR-100 | accuracy: 92.34 -> 92.34 | No clear effect |
| add bypass network without tuners (ablation) | Res-Tuning-Bypass · CIFAR-100 | accuracy: 85.95 -> 86.34 | Helped |
| add horizontal and vertical tuners to the bypass (ablation) | Res-Tuning-Bypass · CIFAR-100 | accuracy: 86.34 -> 89.33 | Helped |
| increase tuner dimension from 10 to 100 (ablation) | Res-Tuning · CIFAR-100 | accuracy: 93.25 -> 92.48 | Hurt |
| increase tuner dimension from 10 to 100 (ablation) | Res-Tuning-Bypass · CIFAR-100 | accuracy: 89.33 -> 89.52 | No clear effect |
| use fewer tuned block layers (1 to 1 instead of 1 to 12) (ablation) | Res-Tuning · CIFAR-100 | accuracy: 93.25 -> 92.23 | Hurt |
| drop layers 1-11 from tuning (ablation) | Res-Tuning · CIFAR-100 | accuracy: 93.25 -> 87.77 | Hurt |
| use Res-Tuning-Bypass instead of full fine-tuning on SD v1.5 | Stable Diffusion v1.5 · COCO2017 | FID: 14.85 -> 14.89 | No clear effect |
| use Res-Tuning instead of full fine-tuning | Stable Diffusion v1.5 · COCO2017 | FID: 14.85 -> 13.96 | Helped |
Closest papers by meaning
- Faster Parameter-Efficient Tuning with Token Redundancy Reduction (CVPR 2025 · similarity 0.86)
- RoSA: Rotational Sparse Adaptation for Memory-Efficient Fine-Tuning (NeurIPS 2026 · similarity 0.85)
- Reducing Fine-Tuning Memory Overhead by Approximate and Memory-Sharing Backpropagation (ICML 2024 · similarity 0.85)
- Sheared Backpropagation for Fine-tuning Foundation Models (CVPR 2024 · similarity 0.85)
- Fundamental Limits of Prompt Tuning Transformers: Universality, Capacity and Efficiency (ICLR 2025 · similarity 0.85)
- Second-Order Fine-Tuning without Pain for LLMs: A Hessian Informed Zeroth-Order Optimizer (ICLR 2025 · similarity 0.84)
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.