AI Paper Atlas

Research map / Large language models / Language, safety and interpretability

Distinguishable Deletion: Unifying Knowledge Erasure and Refusal for Large Language Model Unlearning

Puning Yang, Junchi Yu, Qizhou Wang, Philip Torr, Bo Han, Xiuying Chen

ICML 2026 · Large language models › Language, safety and interpretability · approach: Safety alignment of LLMs · 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
replace token-suppression KD methods with energy-bounded unlearning (EUA)best KD baseline on LLaMA-3.2-3B · TOFU Forget 10%Agg.: 6.967 -> 8.665Improved
add energy-based refusal and energy-boundary training over DR baselinesGuardnail on LLaMA-3.1-8B · TOFU Forget 10%, after jailbreak attackDetection accuracy: 95.6% -> 100.0%Improved
use EUA instead of unmodified model with DR promptingDR methods on LLaMA-3.1-8B · TOFU Forget 10%Retention Quality: 7.456 -> 7.433Hurt
use self-preferenced energy margins instead of manual margins (ablation)EUA on LLaMA-3.2-3B · TOFU Forget 10%Agg.: 7.944 -> 8.665Improved
use self-preferenced energy margins instead of manual margins (ablation)EUA on Qwen2.5-3B · TOFU Forget 10%Erasing Quality: 7.652 -> 9.528Improved
vary top-k used for energy and threshold at inference (ablation)EUA · TOFUdecision reliabilityNo clear effect
add EUA over SatImpSatImp on MUSE-Books · MUSE-Books (ICLM-7B)UtilPres: 45.77 -> 50.45Improved
use EUA instead of SatImp for Bio unlearningSatImp on Zephyr-7B-beta · WMDP Bioaccuracy (lower is better): 0.2545 -> 0.2567No clear effect
EUA under cross-lingual, relearning and jailbreak attacksEUA on LLaMA-3.2-3B · TOFUEQ and LQNo clear effect

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