Research map / Architectures and efficiency / Pruning and efficient tuning
Basis-Oriented Low-rank Transfer for Few-Shot and Test-Time Adaptation
Junghwan Park, Woojin Cho, Junhyuk Heo, Darongsae Kwon, Kookjin Lee
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 |
|---|---|---|---|
| use BOLT instead of aTLAS | few-shot adaptation · General-domain few-shot, 1-shot | accuracy (%): 66.34 -> 71.30 | Improved |
| use BOLT instead of aTLAS | few-shot adaptation · General-domain few-shot, 16-shot | accuracy (%): 74.44 -> 78.46 | Improved |
| use BOLT instead of LoRA | few-shot adaptation · Remote-sensing few-shot, 1-shot | accuracy (%): 70.12 -> 80.77 | Improved |
| use BOLT instead of LoRA | few-shot adaptation · Remote-sensing few-shot, 16-shot | accuracy (%): 90.76 -> 91.86 | Improved |
| use BOLT instead of LoRA | OOD robustness · ImageNet-A, 16-shot | accuracy (%): 10.28 -> 15.88 | Improved |
| use BOLT instead of aTLAS | OOD robustness · ImageNet-R, 16-shot | accuracy (%): 52.21 -> 53.85 | Improved |
| use BOLT instead of LayerNorm tuning | test-time adaptation · CLIP ViT-B/32 general avg | accuracy (%): 62.68 -> 67.74 | Improved |
| use BOLT instead of LayerNorm tuning | test-time adaptation · CLIP ViT-L/14 general avg | accuracy (%): 74.08 -> 74.11 | No clear effect |
| add layer-wise rank r up to about 12 (ablation) | BOLT · CLIP backbones, 16-shot | accuracy | Improved |
| add more source task vectors to the basis (ablation) | BOLT · CLIP backbones, 16-shot | accuracy | Improved |
| use LP++ instead of BOLT at 1-shot | few-shot adaptation · General-domain few-shot, 1-shot | accuracy (%): 71.30 -> 43.34 | Hurt |
Closest papers by meaning
- SDS-LoRA: Overcoming Anisotropic Gradient Scaling in Low-Rank Adaptation (NeurIPS 2026 · similarity 0.87)
- C-LoRA: Continual Low-Rank Adaptation for Pre-trained Visual Models (NeurIPS 2026 · similarity 0.87)
- The Primacy of Magnitude in Low-Rank Adaptation (NeurIPS 2025 · similarity 0.86)
- Bridging The Gap between Low-rank and Orthogonal Adaptation via Householder Reflection Adaptation (NeurIPS 2024 · similarity 0.86)
- Provable Meta-Learning with Low-Rank Adaptations (NeurIPS 2025 · similarity 0.86)
- LOFT: Low-Rank Orthogonal Fine-Tuning via Task-Aware Support Selection (NeurIPS 2026 · 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.
