AI Paper Atlas

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

ICCV 2025 · Continual learning › Continual fine-tuning of LLMs · approach: Synthetic data and merging for CLIP · 0 citations (OpenAlex)

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 changedWhereMeasuredEffect
add Lark (localization + rank-one editing)OrCo-ViT · CIFAR100average accuracy (%): 82.08 -> 86.36Improved
add Lark (localization + rank-one editing)CLOSER-ViT · CIFAR100average accuracy (%): 83.19 -> 86.03Improved
add LarkCLOSER-ViT · mini-ImageNetaverage accuracy (%): 84.11 -> 88.15Improved
add LarkOrCo-ViT · mini-ImageNetaverage accuracy (%): 80.82 -> 86.15Improved
add LarkCLOSER-ViT · CUB200average accuracy (%): 78.79 -> 82.00Improved
add LarkOrCo-ViT · CUB200average accuracy (%): 78.58 -> 80.28Improved
unfreeze backbone and fine-tune without localization (ablation)OrCo-ViT · CIFAR100average accuracy (%): 82.08 -> 77.33Hurt
add localization to unfrozen fine-tuning (ablation)OrCo-ViT · CIFAR100average accuracy (%): 77.33 -> 81.07Improved
add low-rank Rank-All update after localization (ablation)OrCo-ViT · CIFAR100average accuracy (%): 81.07 -> 84.38Improved
use rank-one SVD instead of Rank-All averaging (ablation)OrCo-ViT · CIFAR100average accuracy (%): 84.38 -> 86.36Improved
edit only V instead of Q (ablation)Lark in CLOSER-ViT · CIFAR100average accuracy (%): 84.45 -> 86.03Improved
edit only V instead of Q,K,V jointly (ablation)Lark in CLOSER-ViT · CIFAR100average accuracy (%): 84.93 -> 86.03Improved

Closest papers by meaning

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.

Open scime draft

Get the next scime Dispatch and a monthly field update

Free. We send a confirmation link first; every email has a one-click unsubscribe. Privacy