AI Paper Atlas

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

CLUE: Conflict-guided Localization for LLM Unlearning Framework

Hang Chen, Jiaying Zhu, Xinyu Yang, Wenya Wang

ICLR 2026 · Continual learning › Continual fine-tuning of LLMs · approach: Machine unlearning by selective forgetting

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 conflict-mask stage with forget+retain loss (vs replacing it with full mask) (ablation)CLUE · WMDP Cyber, SST-2 retainretain utility: down 0.264Improved
add forget-neuron fine-tuning stage (ablation)CLUE · WMDP Cyber, SST-2 retainforget efficacy: down 0.045Improved
fine-tune conflict neurons before forget neurons (reverse order) (ablation)CLUE · WMDP Cyber, SST-2 retainretain utility: drop 0.192Hurt
use GA instead of PO as forget loss in both stages (ablation)CLUE · WMDP Cyber, SST-2 retainretain utility: drop 0.529Hurt
use NPO instead of PO as forget loss in both stages (ablation)CLUE · WMDP Cyber, SST-2 retainretain utility: drop 0.093Hurt
use CLUE localization instead of WAGLEWAGLE · WMDP Cyber, SST-2 retainretain utility: 0.771 -> 0.91Improved
localize neurons (localization methods) instead of full fine-tuning with POPO · WMDP Cyber, SST-2 retainretain utility: 0.632 -> 0.731 (DEPN)Improved
use CLUE on WMDP Cyber with Winogrande retain set (forget efficacy vs WAGLE)WAGLE · WMDP Cyber, Winogrande retainforget efficacy: 0.702 -> 0.697No clear effect
use MMLU as retain set instead of specific tasks (ablation)CLUE · WMDP Cyberforget efficacy: lower than specific-task retain setsHurt
increase number of specific retain tasks (ablation)CLUE · WMDP Cyberforget efficacy: decreases as tasks increaseHurt
use denser circuit (lower sparsity) (ablation)CLUE · WMDP Cyber/Bio, PKU-SafeRLHF, SST-2 retainforget efficacy: increases as sparsity decreases from 0.95 to 0.2Improved

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