AI Paper Atlas

Research map / Adversarial robustness and security / Attacks on text and graphs

LLM Unlearning via Neural Activation Redirection

William F. Shen, Xinchi Qiu, Meghdad Kurmanji, Alex Iacob, Lorenzo Sani, Yihong Chen, Nicola Cancedda, Nicholas D. Lane

NeurIPS 2025 · Adversarial robustness and security › Attacks on text and graphs · approach: Adversarial prompts for language models · 1 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 DPO with LUNARLlama2-7B · PISTOLDeviation Score (DS): 22.8 -> 7.8Improved
Replace DPO with LUNARLlama2-7B · PISTOLControl Score: 0.524 -> 0.677Improved
Replace DPO with LUNARGemma-7B · PISTOLDeviation Score (DS): 23.4 -> 6.3Improved
Replace DPO with LUNARGemma-7B · PISTOLControl Score: 0.692 -> 0.701No clear effect
Replace DPO with LUNARQwen2-7B · PISTOLDeviation Score (DS): 24.6 -> 5.9Improved
Replace DPO with LUNARQwen2-7B · PISTOLControl Score: 0.594 -> 0.640Improved
Replace DPO with LUNARLlama2-7B · TOFUDeviation Score (DS): 15.2 -> 14.9Improved
Replace DPO with LUNARLlama2-7B · TOFUControl Score: 0.515 -> 0.608Improved
Replace DPO with LUNARGemma-7B · TOFUDeviation Score (DS): 20.2 -> 13.1Improved
Replace DPO with LUNARGemma-7B · TOFUControl Score: 0.588 -> 0.659Improved
Replace NPO with LUNARQwen2-7B · TOFUDeviation Score (DS): 26.7 -> 14.3Improved
Replace NPO with LUNARQwen2-7B · TOFUControl Score: 0.477 -> 0.609Improved

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