**摘要**
LLMs memorize sensitive training data, including personally identifiable information (PII), creating a pressing need for reliable post hoc removal methods. Unlearning has emerged as a promising solution, with state-of-the-art(SOTA) methods often following a localize-first, unlearn-second paradigm that targets specific model parameters. However, existing benchmarks evaluate unlearning solely at the
👤 作者: Matteo Boglioni, Thibault Rousset, Siva Reddy, 马吕斯·莫斯巴赫, Verna Dankers

---
🔗 **[LACUNA ： LLM Unlearning本地化精度评估测试平台](https://arxiv.org/abs/2607.02513v1)**

> LACUNA: A Testbed for Evaluating Localization Precision for LLM Unlearning
🏷️ 来源: ArXiv cs.AI
⏱️ 2026-07-03 14:00