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
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 changed | Where | Measured | Effect |
|---|---|---|---|
| Replace DPO with LUNAR | Llama2-7B · PISTOL | Deviation Score (DS): 22.8 -> 7.8 | Improved |
| Replace DPO with LUNAR | Llama2-7B · PISTOL | Control Score: 0.524 -> 0.677 | Improved |
| Replace DPO with LUNAR | Gemma-7B · PISTOL | Deviation Score (DS): 23.4 -> 6.3 | Improved |
| Replace DPO with LUNAR | Gemma-7B · PISTOL | Control Score: 0.692 -> 0.701 | No clear effect |
| Replace DPO with LUNAR | Qwen2-7B · PISTOL | Deviation Score (DS): 24.6 -> 5.9 | Improved |
| Replace DPO with LUNAR | Qwen2-7B · PISTOL | Control Score: 0.594 -> 0.640 | Improved |
| Replace DPO with LUNAR | Llama2-7B · TOFU | Deviation Score (DS): 15.2 -> 14.9 | Improved |
| Replace DPO with LUNAR | Llama2-7B · TOFU | Control Score: 0.515 -> 0.608 | Improved |
| Replace DPO with LUNAR | Gemma-7B · TOFU | Deviation Score (DS): 20.2 -> 13.1 | Improved |
| Replace DPO with LUNAR | Gemma-7B · TOFU | Control Score: 0.588 -> 0.659 | Improved |
| Replace NPO with LUNAR | Qwen2-7B · TOFU | Deviation Score (DS): 26.7 -> 14.3 | Improved |
| Replace NPO with LUNAR | Qwen2-7B · TOFU | Control Score: 0.477 -> 0.609 | Improved |
Closest papers by meaning
- Keeping an Eye on LLM Unlearning: The Hidden Risk and Remedy (NeurIPS 2025 · similarity 0.91)
- Towards Robust Knowledge Unlearning: An Adversarial Framework for Assessing and Improving Unlearning Robustness in Large Language Models (AAAI 2025 · similarity 0.91)
- LLM Unlearning via Loss Adjustment with Only Forget Data (ICLR 2025 · similarity 0.90)
- Towards LLM Unlearning Resilient to Relearning Attacks: A Sharpness-Aware Minimization Perspective and Beyond (ICML 2025 · similarity 0.90)
- Unlearning or Obfuscating? Jogging the Memory of Unlearned LLMs via Benign Relearning (ICLR 2025 · similarity 0.90)
- Invariance Makes LLM Unlearning Resilient Even to Unanticipated Downstream Fine-Tuning (ICML 2025 · similarity 0.90)
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.
