AI Paper Atlas

Research map / Continual learning / Continual fine-tuning of LLMs

Merge to Remember: Sharpness-Aware Isotropic Merging for Continual Learning

Qun Yang, Enneng Yang, Wei Chen, Li Shen, Long Lan

ICML 2026 · Continual learning › Continual fine-tuning of LLMs · approach: Low-rank and model-merging adaptation

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
use sharpness-aware perturbation instead of random noise in SA-BCD (ablation)SA-BCD · ViT-B/32average accuracy (%): 73.20 -> 82.74Improved
add sharpness-aware perturbation (vs none) (ablation)SA-BCD · ViT-B/32average accuracy (%): 77.71 -> 82.74Improved
select parameters by momentum instead of instantaneous gradient (ablation)SA-BCD · ViT-B/32average accuracy (%): 81.02 -> 82.74Improved
add block selection of top 30% parameters (vs update all) (ablation)SA-BCD · ViT-B/32average accuracy (%): 79.54 -> 82.74Improved
use SA-BCD instead of AdamW fine-tuning (ablation)SAIM · ViT-B/32 20 tasksloss drift after 8 tasks: 1.4522 -> 0.0407Improved
add SA-BCD fine-tuning to TIES-Merging (ablation)TIES-Merging · ViT-B/32 20 tasksaccuracy (%): 59.9 -> 63.4Improved
add SA-BCD fine-tuning to Task Arithmetic (ablation)Task Arithmetic · ViT-B/32 20 tasksaccuracy (%): 60.0 -> 61.2Improved
add SA-BCD fine-tuning to simple averaging (SWA) (ablation)SWA · ViT-B/32 20 tasksaccuracy (%): 61.1 -> 59.6Hurt
add SA-BCD fine-tuning to SAIM merging (ablation)SAIM · ViT-B/32 20 tasksaccuracy (%): 70.8 -> 73.6Improved
use SAIM instead of EWC/ER at few tasksSAIM · ViT-B/32 8 tasksaverage accuracy (%): 84.7 (EWC) -> 82.1Hurt
use SAIM instead of OPCMSAIM · ViT-L/14 20 tasksaverage accuracy (%): 76.0 -> 87.5Improved
use SAIM instead of best PEFT method (SD-LoRA)SAIM · ViT-B/32 20 tasksaverage accuracy (%): 66.4 -> 73.6Improved

Earlier methods this paper compares with

Named baselines from earlier accepted papers in this paper's results tables: better than 2 of 2 on most 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

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