[Paper] Maximizing Local Entropy Where It Matters: Prefix-Aware Localized LLM Unlearning
Machine unlearning aims to forget sensitive knowledge from Large Language Models (LLMs) while maintaining general utility. However, existing approaches typicall...