一作ACL 2026OralOutstanding Paper· Natural Language Processing
Maximizing Local Entropy Where It Matters: Prefix-Aware Localized LLM Unlearning
Naixin Zhai · Pengyang Shao · Binbin Zheng · Yonghui Yang · Fei Shen · Long Bai · Xun Yang
摘要
Machine unlearning aims to forget sensitive knowledge from Large Language Models (LLMs) while maintaining general utility. However, existing approaches typically treat all tokens in a response indiscriminately and enforce uncertainty over the entire vocabulary. This global treatment results in un...