知识卡片

普通读书笔记卡 · 8864.a · 生成模型:GPT-6

隐藏表示可形成非线性边界

理解等级:中级

多个仿射响应经ReLU组合后,可在原始输入空间形成折线等非线性边界,从而分开单条直线无法分开的样本。手工指定权重展示的是某组参数存在,不等于训练算法一定能找到它,也不证明所有非线性分类都由同样的小网络解决。

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 example demonstrates that is possible to solve nonlinear classification problems with a deep net. In practice, we would want to learn the parameter settings that achieve this classification. One way to do so would be to enumerate all possible parameter settings and pick one that successfully separates the zeros from the ones. This kind of exhaustive enumeration is a slow process, but don't worry,

    参考来源

    - 位置:《Foundations of Computer Vision (Antonio Torralba, Phillip Isola etc.)》第12章《神经网络》(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/c012.xhtml) - 结论依据:摘录中关于隐藏表示可形成非线性边界的定义、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 example demonstrates that is possible to solve nonlinear classification problems with a deep net. In practice, we would want to learn the parameter settings that achieve this classification. One way to do so would be to enumerate all possible parameter settings and pick one that successfully separates the zeros from the ones. This kind of exhaustive enumeration is a slow process, but don't worry,