知识卡片

普通读书笔记卡 · 8872 · 生成模型:GPT-6

逐点非线性反传就是逐项门控

理解等级:中级

逐点层的雅可比为对角矩阵,因此反传可直接写δin=δout⊙h′(xin),不必构造矩阵。ReLU使负输入位置的梯度为零,正输入位置保留;零点采用哪条规则需明确。反向操作是乘局部导数,不是再次对梯度应用ReLU。

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

      with a = h′(xin[0]), b = h′(xin[1]), and c = h′(xin[2]). We can simplify this equation as follows: (14.33) The full set of operations for a pointwise layer is shown next in figure 14.13. Figure 14.13: Pointwise layer forward and backward 14.6.3 Backpropagation for Loss Layers The last layer we need to define for a complete MLP is the loss layer. As a simple example,

    参考来源

    - 位置:《Foundations of Computer Vision (Antonio Torralba, Phillip Isola etc.)》第14章《反向传播》(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/c014.xhtml) - 结论依据:摘录中关于逐点非线性反传就是逐项门控的定义、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:with a = h′(xin[0]), b = h′(xin[1]), and c = h′(xin[2]). We can simplify this equation as follows: (14.33) The full set of operations for a pointwise layer is shown next in figure 14.13. Figure 14.13: Pointwise layer forward and backward 14.6.3 Backpropagation for Loss Layers The last layer we need to define for a complete MLP is the loss layer. As a simple example,