← 返回刚才阅读的位置

知识卡片

普通读书笔记卡 · 9004.c · 生成模型:GPT-6

公开查询不等于梯度访问

理解等级:高级

基于梯度直接优化扰动通常需要模型或梯度信息;提供公共预测接口并不自动暴露架构和梯度。黑盒也可能通过查询或迁移构造攻击,但需要另行说明信息条件和预算。解释风险时须区分威胁模型,不能把白盒示例的条件直接套到所有 API。

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

      This type of attack required that the attacker has access to your system (the neural net f) so that they can optimize the noise pattern ϵ to fool the system. Unfortunately, if you want to deploy your system to the public, then you have to give people access to it and that makes it vulnerable to attack.

    参考来源

    - 位置:《Foundations of Computer Vision (Antonio Torralba, Phillip Isola etc.)》第35章《数据偏差与分布漂移》(来源项目:a19c7933-4eda-4176-aebd-30fe7e1ad292;源文件:extracted/Foundations of Computer Vision (Antonio Torralba, Phillip Isola etc.)/_epub-src/OPS/xhtml/c035b.xhtml) - 结论依据:摘录中关于公开查询不等于梯度访问的定义、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:This type of attack required that the attacker has access to your system (the neural net f) so that they can optimize the noise pattern ϵ to fool the system. Unfortunately, if you want to deploy your system to the public, then you have to give people access to it and that makes it vulnerable to attack.