Research map / Continual learning / Continual fine-tuning of LLMs
PLAN: Proactive Low-Rank Allocation for Continual Learning
Xiequn Wang, Zhan Zhuang, Yu Zhang
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 basis selection mechanism for A_t (ablation) | PLAN · ImageNet-R (N=20) | Acc: 68.35 -> 71.06 | Improved |
| add perturbation-based min-max optimization of B_t (ablation) | PLAN · ImageNet-R (N=20) | Acc: 70.65 -> 71.06 | Improved |
| add selection mechanism (ablation) | PLAN · ImageNet-R (N=5) | Acc: 75.97 -> 77.79 | Improved |
| add perturbation optimization (ablation) | PLAN · ImageNet-R (N=10) | Acc: 74.97 -> 75.25 | Improved |
| add both components over Inc-LoRA (ablation) | Inc-LoRA · ImageNet-R (N=20) | Acc: 52.12 -> 71.06 | Improved |
| use standard basis instead of random orthogonal basis (ablation) | PLAN · CIFAR-100 | Acc: 81.21 -> 87.54 | Improved |
| use standard basis instead of LoRA-GA basis (ablation) | PLAN · CIFAR-100 | Acc: 84.30 -> 87.54 | Improved |
| use p=2 instead of p=1 (ablation) | PLAN · CIFAR-100 | Acc: 80.32 -> 87.54 | Improved |
| use p=2 instead of p=infinity (ablation) | PLAN · CIFAR-100 | Acc: 79.94 -> 87.54 | Improved |
| set rho=0.01 instead of 0.1 (ablation) | PLAN · ImageNet-R (N=5) | Acc: 75.23 -> 77.79 | Improved |
| set rho=0.01 instead of 0.001 (ablation) | PLAN · ImageNet-R (N=5) | Acc: 76.38 -> 77.79 | Improved |
| PLAN vs InfLoRA with iBOT pretraining on ImageNet-R N=20 | PLAN · ImageNet-R N=20 (iBOT-1k) | Acc: 78.11 -> 78.39 | Mixed |
Earlier methods this paper compares with
Named baselines from earlier accepted papers in this paper's results tables: better than 1 of 1 on most shared results.
- InfLoRA — InfLoRA: Interference-Free Low-Rank Adaptation for Continual Learning (CVPR 2024) · better on 7 of 8 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
- C-LoRA: Continual Low-Rank Adaptation for Pre-trained Visual Models (NeurIPS 2026 · similarity 0.90)
- Learn more, but bother less: parameter efficient continual learning (NeurIPS 2024 · similarity 0.90)
- PaLoRA: Paced Low-Rank Adaptation for Continual Learning (NeurIPS 2026 · similarity 0.90)
- CL-LoRA: Continual Low-Rank Adaptation for Rehearsal-Free Class-Incremental Learning (CVPR 2025 · similarity 0.90)
- InfLoRA: Interference-Free Low-Rank Adaptation for Continual Learning (CVPR 2024 · similarity 0.90)
- Continuous Subspace Optimization for Continual Learning (NeurIPS 2025 · similarity 0.89)
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.
