Research map / Continual learning / Continual fine-tuning of LLMs
VD-LoRA: Adaptive Reuse of Low-Rank Directions for Continual Learning
Luqiong Ding, Jiayao Tan, Chenggong Ni, Fuyuan Hu, Fan Lyu
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 recursive KL regularization (RKL) (ablation) | VD-LoRA · ImageNet-R (ablation) | AAA (N=5): 80.22 -> 83.30 | Improved |
| add recursive KL regularization (RKL) (ablation) | VD-LoRA · ImageNet-R (ablation) | AAA (N=10): 76.54 -> 80.20 | Improved |
| add recursive KL regularization (RKL) (ablation) | VD-LoRA · ImageNet-R (ablation) | AAA (N=20): 68.95 -> 73.80 | Improved |
| add uncertainty-guided direction selection (SEL) (ablation) | VD-LoRA · ImageNet-R (ablation) | AAA (N=20): 73.80 -> 82.60 | Improved |
| add uncertainty-guided direction selection (SEL) (ablation) | VD-LoRA · ImageNet-R (ablation) | AAA (N=10): 80.20 -> 83.60 | Improved |
| add uncertainty-guided direction selection (SEL) (ablation) | VD-LoRA · ImageNet-R (ablation) | AAA (N=5): 83.30 -> 85.10 | Improved |
| add stability regularization (STR) (ablation) | VD-LoRA · ImageNet-R (ablation) | AAA (N=20): 82.60 -> 83.90 | Improved |
| add stability regularization (STR) (ablation) | VD-LoRA · ImageNet-R (ablation) | AAA (N=10): 83.60 -> 84.97 | Improved |
| add stability regularization (STR) (ablation) | VD-LoRA · ImageNet-R (ablation) | AAA (N=5): 85.10 -> 85.77 | Improved |
| raise rank budget to r=10 from r=1 (ImageNet-R T=20) (ablation) | VD-LoRA · ImageNet-R (N=20, rank sweep) | accuracy: 75.92 -> 77.53 | Improved |
| raise rank budget to r=50 from r=10 (ablation) | VD-LoRA · ImageNet-R (N=20, rank sweep) | accuracy: 77.53 -> 76.34 | Hurt |
| vary gamma in uncertainty-weighted selection from 0 to 0.5 (ablation) | VD-LoRA · CIFAR-100 | accuracy: 89.10 -> 89.69 | Improved |
Earlier methods this paper compares with
Named baselines from earlier accepted papers in this paper's results tables: better than 3 of 3 on most shared results.
- CL-LoRA — CL-LoRA: Continual Low-Rank Adaptation for Rehearsal-Free Class-Incremental Learning (CVPR 2025) · better on 8 of 8 shared results
- InfLoRA — InfLoRA: Interference-Free Low-Rank Adaptation for Continual Learning (CVPR 2024) · better on 5 of 5 shared results
- SD-LoRA — SD-LoRA: Scalable Decoupled Low-Rank Adaptation for Class Incremental Learning (ICLR 2025) · better on 3 of 3 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
- SD-LoRA: Scalable Decoupled Low-Rank Adaptation for Class Incremental Learning (ICLR 2025 · similarity 0.92)
- CL-LoRA: Continual Low-Rank Adaptation for Rehearsal-Free Class-Incremental Learning (CVPR 2025 · similarity 0.91)
- PaLoRA: Paced Low-Rank Adaptation for Continual Learning (NeurIPS 2026 · similarity 0.90)
- C-LoRA: Continual Low-Rank Adaptation for Pre-trained Visual Models (NeurIPS 2026 · similarity 0.89)
- LoRA Subtraction for Drift-Resistant Space in Exemplar-Free Continual Learning (CVPR 2025 · similarity 0.89)
- AuroRA: Breaking Low-Rank Bottleneck of LoRA with Nonlinear Mapping (NeurIPS 2025 · similarity 0.88)
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.
