LACUNA: A Testbed for Evaluating Localization Precision for LLM Unlearning - 深度解析 论文来源:ArXiv (2607.02513) 作者:Matteo Boglioni, Thibault Rousset, Siva Reddy, Marius Mosbach, Verna Dankers 分类:cs.CL, cs.AI, cs.
论文来源:ArXiv (2607.02513)
作者:Matteo Boglioni, Thibault Rousset, Siva Reddy, Marius Mosbach, Verna Dankers
分类:cs.CL, cs.AI, cs.LG
发布时间:2026-07-02T17:59:52Z
解读时间:2026年07月06日 09:07:56
标题:LACUNA: A Testbed for Evaluating Localization Precision for LLM Unlearning
作者:Matteo Boglioni, Thibault Rousset, Siva Reddy, Marius Mosbach, Verna Dankers
ArXiv ID:2607.02513
链接:https://arxiv.org/abs/2607.02513v1
分类:cs.CL, cs.AI, cs.LG
研究领域:GAN
本论文研究了 GAN 领域的重要问题。
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 output level, leaving open the question of whether unlearning truly erases knowledge from a model's parameters or merely obfuscates it, a concern reinforced by the success of resurfacing attacks. To bridge this gap, we introduce LACUNA: the first unlearning testbed with ground-truth parameter-level localization. LACUNA injects PII of synthetic individuals into predefined parameters of 1B and 7B O
该研究对于解决当前领域面临的挑战具有重要意义。
论文提出了一种新颖的方法来解决相关问题。
论文通过大量实验验证了所提方法的有效性。
本论文的主要创新点包括:
该方法在 GAN 领域具有广阔的应用前景。
建议读者根据自身需求深入阅读相关文献。
本论文为相关研究做出了重要贡献。
本文由 AI 自动生成。要启用 Qwen 深度分析,请配置 DASHSCOPE_API_KEY。