arXiv · cs.AI · Wenxi Dai, Jingyu Hu, Yanting Yang, Mengnan Du · 10 小时前模型评分42
从表征到读出:语言模型中成员信号的机制分析
From Representation to Readout: Mechanistic Analysis of Membership Signals in Language Models
导读
研究对大语言模型(LLM,Large Language Model)中的成员信号做了机制分析,提出可解码信息分数、已用信息分数与读出可达性差距(DUG)三项互补指标,分别量化内部表征中可解码的成员信息及其通过模型读出暴露的程度。
分类置信度
中 · 待人工复核
- 只有单一媒体来源,需要人工复核。
来源级别:媒体与个人
事件类型:研究发布
证据摘录
- “We introduce Decodable Information Score, Used Information Score, and Readout Accessibility Gap (DUG), a set of complementary measures that separately quantify membership information decodable from internal representations and the extent to which it is exposed through the model's readout.”
来源:arXiv · cs.AI · arxiv.org