Research map / Architectures and efficiency / Pruning and efficient tuning
Efficient ViT adaptation
Adapts pre-trained vision transformers with low-rank and factorized parameter-efficient tuning.
62 accepted papers at ICML, NeurIPS, ICLR, CVPR and AAAI (2019–2026) · an approach in Pruning and efficient tuning, keyword cluster “fine tuning · lora · adaptation”
State History Tried Checklist Preprints Who Export
Where it stands
Not enough benchmark data Fewer than four papers report the same benchmark, so there is no record to follow.
35 papers in 2023–24 and 16 in 2025–26 (at ICML, NeurIPS, ICLR, CVPR and AAAI); its share of the topic shrank, from 12.6% to 2.9%.
How this is decided
A benchmark is tracked when at least four papers of this approach report their own method on the same dataset and metric in at least two different years. Its record is the best number reported up to 2024 compared with the best reported in 2025–26: still improving means at least 1% better, flat means at most 0.2% better with at least two papers from 2025–26 trying. Numbers on a different scale from the rest (a fraction among percentages) and far outliers are left out. Results come from different papers with different setups, so they are signals, not head-to-head comparisons. The share of papers is counted at ICML, NeurIPS, ICLR, CVPR and AAAI.
How it got here
Most cited paper of each year
- 2019 · Efficient Augmentation via Data Subsampling (ICLR 2019 · 12 citations)
- 2022 · AdaptFormer: Adapting Vision Transformers for Scalable Visual Recognition (NeurIPS 2022 · 262 citations)
- 2023 · FacT: Factor-Tuning for Lightweight Adaptation on Vision Transformer (AAAI 2023 · 98 citations)
- 2024 · VMT-Adapter: Parameter-Efficient Transfer Learning for Multi-Task Dense Scene Understanding (AAAI 2024 · 62 citations)
- 2025 · TinyFusion: Diffusion Transformers Learned Shallow (CVPR 2025 · 8 citations)
- 2026 · Small but Mighty: Dynamic Wavelet Expert-Guided Fine-Tuning of Large-Scale Models for Optical Remote Sensing Object Segmentation (AAAI 2026 · 3 citations)
Earliest papers here
- Efficient Augmentation via Data Subsampling (ICLR 2019 · 12 citations)
- AdaptFormer: Adapting Vision Transformers for Scalable Visual Recognition (NeurIPS 2022 · 262 citations)
- Pushing the Limits of Simple Pipelines for Few-Shot Learning: External Data and Fine-Tuning Make a Difference (CVPR 2022 · 178 citations)
What has been tried
53 papers here report 630 measured changes (a component added, removed or swapped, compared with a baseline): helped 62%, no clear effect 9%, hurt 26%, mixed 3%. Papers mostly publish changes that helped, so "helped" is over-represented.
Ablations in recent papers
| What was changed | Measured on | Result | Effect | Paper |
|---|---|---|---|---|
| set down-projection ratio rho=1 | PASCAL-Context four-task average · relative improvement vs. Single-task Full Fine-tuning | 4.27% | Helped | VMT-Adapter: Parameter-Efficient Transfer Learning for Multi-Task Dense Scene Understanding (AAAI 2024 · 62 citations) |
| replace random token dropping with WPR importance scoring | ImageNet · Acc@1 | 76.8% (-3.0%) -> 78.6% (+1.8%) | Helped | Zero-TPrune: Zero-Shot Token Pruning through Leveraging of the Attention Graph in Pre-Trained Transformers (CVPR 2024 · 39 citations) |
| Add the input-side equilibrium layer | CIFAR10 · accuracy | 0.8501 -> 0.8628 | Helped | LION: Implicit Vision Prompt Tuning (AAAI 2024 · 26 citations) |
| Add a shared global DWConv after Swish in CSN | VTAB-1K · Average accuracy | 76.7 -> 77.7 | Helped | DTL: Disentangled Transfer Learning for Visual Recognition (AAAI 2024 · 21 citations) |
| Add the residual term to left-and-right scaling | FGVC · mean accuracy | 88.9 -> 90.4 | Helped | Low-Rank Rescaled Vision Transformer Fine-Tuning: A Residual Design Approach (CVPR 2024 · 20 citations) |
| Choose rank per task from [1..32] instead of using a fixed rank | VTAB test sets · average accuracy | 77.6 -> 77.9 | Helped | Adapters Strike Back (CVPR 2024 · 12 citations) |
| Add MaskedKD during TinyDiT-D14 recovery | ImageNet 256×256 · FID | 5.73 -> 3.73 | Helped | TinyFusion: Diffusion Transformers Learned Shallow (CVPR 2025 · 8 citations) |
| Use DyT with token dispatching before MLP rather than attention dispatch | CIFAR-100 · accuracy | 84.58 -> 91.37 | Helped | Dynamic Tuning Towards Parameter and Inference Efficiency for ViT Adaptation (NeurIPS 2024 · 6 citations) |
| add task prompts and the TPC-operator without DTF | PASCAL-Context · relative improvement over single-task fine-tuning | +1.3 -> +1.79 (+0.49) | Helped | TADFormer: Task-Adaptive Dynamic TransFormer for Efficient Multi-Task Learning (CVPR 2025 · 4 citations) |
| Replace LoRA with HTA on Wq and Wv | VTAB-1k · Mean Total accuracy | 72.3 -> 72.7 | Helped | Efficient Adaptation of Pre-trained Vision Transformer via Householder Transformation (NeurIPS 2024 · 3 citations) |
| Preserve aspect ratio and high-resolution details with multiscale inputs | AVA · PLCC | 0.734 -> 0.779 | Helped | Charm: The Missing Piece in ViT Fine-Tuning for Image Aesthetic Assessment (CVPR 2025 · 2 citations) |
| use joint singular-value modulation alone | PASCAL-MTL · relative improvement over single-task tuning | 2.77% | Helped | DiTASK: Multi-Task Fine-Tuning with Diffeomorphic Transformations (CVPR 2025 · 2 citations) |
What reviewers will expect
Datasets and baselines named in the 8 papers of this approach from 2025–2026 that have a full-text summary. Leaving out the common ones is a frequent review question.
Datasets
- Food-101 · 3 of 8 papers
- VTAB-1K · 2 of 8 papers
- ImageNet-1K · 2 of 8 papers
- Flowers102 · 2 of 8 papers
- CIFAR100 · 2 of 8 papers
- FGVC-Aircraft · 2 of 8 papers
Baselines
- LoRA · 4 of 8 papers
- BitFit · 3 of 8 papers
- Adapter · 3 of 8 papers
- VPT · 3 of 8 papers
- AdaptFormer · 2 of 8 papers
- NOAH · 2 of 8 papers
- SSF · 2 of 8 papers
- MTLoRA · 2 of 8 papers
Recent preprints
Loading recent arXiv preprints…
Who publishes on it
Institutions on the most papers of this approach since 2023 (of 51 papers; from OpenAlex and the papers' affiliations).
- Tsinghua University · 4 papers
- Nanjing University · 3 papers
- Peking University · 3 papers
- Tencent · 3 papers
- National University of Singapore · 3 papers
- Xi’an University of Architecture and Technology · 3 papers
- Huawei · 2 papers
- Southeast University · 2 papers
- Georgia Institute of Technology · 2 papers
- University of Electronic Science and Technology of China · 2 papers
Use it
Download BibTeX of 34 key papersOpen the topic in the interactive map
Get alerts. Describe your idea in scime atlas and save it (free scime account): you get a weekly email when new preprints land close to it.
Related approaches
Same keyword cluster
- Low-rank compression and merging (226 papers)
- Adapter-based parameter-efficient tuning (149 papers)
- Parameter-efficient transfer analysis (97 papers)
- LoRA variants and theory (121 papers)
Elsewhere in the topic
- Structured network pruning (162 papers)
- Sparse training from scratch (95 papers)
- Layer-wise pruning of LLMs (91 papers)
- Joint pruning and quantization (78 papers)
- Lottery ticket discovery (42 papers)
- Pruning at initialization (24 papers)
Benchmark numbers, ablations and baseline numbers are extracted by AI from each paper's full text and can be wrong; the paper is the authority. Data: results index of 2026-11-02, citation counts from OpenAlex (2026-10-09).