知识卡片

普通读书笔记卡 · 9264.b · 生成模型:GPT-6

联邦训练参数仍可能泄露信息

理解等级:中级

联邦训练可减少集中传送原始记录,但参数或更新本身仍可能暴露训练信息,也不自动解决各端分布差异。隐私机制与模型效果需要分别评价,不能把数据留在本地等同已有完整隐私保证。

这张卡片看不懂?

已有反馈 · 0 条

    还没有人提交反馈。

    在整书笔记中的位置

    原书出处 · 1 处

    • · 查看原书依据

      Currently, the ubiquitous deployment of ML on fleets of autonomous vehicles is limited by two main factors. First, the collected data are often not representative of the environments the autonomous vehicle will encounter. For example, consider the problem of deploying an ML system for a fleet of ships for navigation. The objective of the model is to perform object detection by processing images from the

    参考来源

    - 位置:《Machine Learning Fundamentals: Concepts, Models, and Applications》第10章《现实应用》(来源项目:a19c7933-4eda-4176-aebd-30fe7e1ad292;源文件:extracted/Machine Learning Fundamentals: Concepts, Models, and Applications/_epub-src/OEBPS/xhtml/25_chapter10.xhtml) - 结论依据:摘录中关于联邦训练参数仍可能泄露信息的定义、比较或条件支撑卡片;原书表述不严谨的范围已在覆盖记录逐项限定,不把案例结果外推为普遍结论。 - 原始内容:Currently, the ubiquitous deployment of ML on fleets of autonomous vehicles is limited by two main factors. First, the collected data are often not representative of the environments the autonomous vehicle will encounter. For example, consider the problem of deploying an ML system for a fleet of ships for navigation. The objective of the model is to perform object detection by processing images from the