从语言模型中擦除概念性知识
Erasing Conceptual Knowledge from Language Models
AI总结：
本文提出ELM方法，利用语言模型自身作为分类器进行概念级遗忘，通过低秩更新降低特定概念生成概率，在生物安全、网络安全和文学领域实现了接近随机的擦除效果，同时保持模型通用能力与鲁棒性。
摘要：
在这项工作中，我们提出了语言记忆擦除（Erasure of Language Memory，ELM），这是一种基于原则的概念级遗忘方法，其核心操作是匹配由模型自身内省式分类能力所定义的分布。我们的关键洞察是：有效的遗忘应当利用模型评估自身知识的能力，将语言模型本身用作分类器，以识别并降低生成与不期望概念相关内容的可能性。ELM 应用这一框架来创建有针对性的低秩更新，在降低概念特定内容生成概率的同时，保留模型的更广泛能力。我们在生物安全、网络安全和文学领域擦除任务上展示了 ELM 的有效性。对比评估表明，经过 ELM 修改的模型在针对已擦除概念的评估上达到接近随机的性能，同时保持生成连贯性，在无关任务上维持基准性能，并对对抗性攻击表现出强鲁棒性。我们的代码、数据和训练模型可在 https://elm.baulab.info 获取。
英文摘要：
In this work, we introduce Erasure of Language Memory (ELM), a principled approach to concept-level unlearning that operates by matching distributions defined by the model's own introspective classification capabilities. Our key insight is that effective unlearning should leverage the model's ability to evaluate its own knowledge, using the language model itself as a classifier to identify and reduce the likelihood of generating content related to undesired concepts. ELM applies this framework to create targeted low-rank updates that reduce generation probabilities for concept-specific content while preserving the model's broader capabilities. We demonstrate ELM's efficacy on biosecurity, cybersecurity, and literary domain erasure tasks. Comparative evaluation reveals that ELM-modified models achieve near-random performance on assessments targeting erased concepts, while simultaneously preserving generation coherence, maintaining benchmark performance on unrelated tasks, and exhibiting strong robustness to adversarial attacks. Our code, data, and trained models are available at https://elm.baulab.info
机构、总结、中文摘要均由AI分析而来，可能存在错误，请以论文正文为准。
