研究人员开发了FedAttr，一种新颖的协议，旨在识别联邦学习设置中哪些客户端在微调大型语言模型时使用了带水印的数据。该方法解决了联邦学习中的挑战，在这种学习中，安全聚合通常会掩盖单个客户端的贡献。FedAttr采用配对子集差分机制来估计客户端更新，并采用具有水印检测器的差分评分方法，在经验测试中实现了完美的真阳性率和零假阳性率。 AI
影响 增强了协作式LLM微调场景中的数据所有权和归因能力。
排序理由 这是一篇研究论文，详细介绍了在联邦LLM微调中实现隐私保护归因的新协议。
AI 生成摘要 · Google Gemini · 来自 2 个来源。 我们如何撰写摘要 →
arXiv:2605.06596v1 Announce Type: cross Abstract: Watermark radioactivity testing type of methods can detect whether a model was trained on watermarked documents, and have become key tools for protecting data ownership in the fine-tuning of large language models (LLMs). Existing …
Watermark radioactivity testing type of methods can detect whether a model was trained on watermarked documents, and have become key tools for protecting data ownership in the fine-tuning of large language models (LLMs). Existing works have proved their effectiveness in centraliz…