Research map / Continual learning / Continual fine-tuning of LLMs
Lark: Low-Rank Updates After Knowledge Localization for Few-shot Class-Incremental Learning
Jinxin Shi, Jiabao Zhao, Yifan Yang, Xingjiao Wu, Jiawen Li, Liang He
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 Lark (localization + rank-one editing) | OrCo-ViT · CIFAR100 | average accuracy (%): 82.08 -> 86.36 | Improved |
| add Lark (localization + rank-one editing) | CLOSER-ViT · CIFAR100 | average accuracy (%): 83.19 -> 86.03 | Improved |
| add Lark | CLOSER-ViT · mini-ImageNet | average accuracy (%): 84.11 -> 88.15 | Improved |
| add Lark | OrCo-ViT · mini-ImageNet | average accuracy (%): 80.82 -> 86.15 | Improved |
| add Lark | CLOSER-ViT · CUB200 | average accuracy (%): 78.79 -> 82.00 | Improved |
| add Lark | OrCo-ViT · CUB200 | average accuracy (%): 78.58 -> 80.28 | Improved |
| unfreeze backbone and fine-tune without localization (ablation) | OrCo-ViT · CIFAR100 | average accuracy (%): 82.08 -> 77.33 | Hurt |
| add localization to unfrozen fine-tuning (ablation) | OrCo-ViT · CIFAR100 | average accuracy (%): 77.33 -> 81.07 | Improved |
| add low-rank Rank-All update after localization (ablation) | OrCo-ViT · CIFAR100 | average accuracy (%): 81.07 -> 84.38 | Improved |
| use rank-one SVD instead of Rank-All averaging (ablation) | OrCo-ViT · CIFAR100 | average accuracy (%): 84.38 -> 86.36 | Improved |
| edit only V instead of Q (ablation) | Lark in CLOSER-ViT · CIFAR100 | average accuracy (%): 84.45 -> 86.03 | Improved |
| edit only V instead of Q,K,V jointly (ablation) | Lark in CLOSER-ViT · CIFAR100 | average accuracy (%): 84.93 -> 86.03 | Improved |
Closest papers by meaning
- Tripartite Weight-Space Ensemble for Few-Shot Class-Incremental Learning (CVPR 2025 · similarity 0.87)
- SAME: Sparse and Anchored Model Editing for Heterogeneous Incremental Learning under Limited Data (CVPR 2026 · similarity 0.86)
- Warping the Space: Weight Space Rotation for Class-Incremental Few-Shot Learning (ICLR 2023 · similarity 0.86)
- Pre-trained Vision and Language Transformers Are Few-Shot Incremental Learners (CVPR 2024 · similarity 0.85)
- C-LoRA: Continual Low-Rank Adaptation for Pre-trained Visual Models (NeurIPS 2026 · similarity 0.84)
- SD-LoRA: Scalable Decoupled Low-Rank Adaptation for Class Incremental Learning (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.
