知识卡片
雅可比形状取决于导数布局
本书把输入向量写成N×1,标量对向量的导数写为1×N行向量,M维输出对N维输入的雅可比为M×N。每行对应一个输出,每列对应一个输入。这是布局约定,换用列梯度资料或自动微分接口时,必须核对形状而非机械复制链式相乘顺序。
这张卡片看不懂?
这张卡片已通过人工审查,不再接受失败反馈。
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
If y is an M-dimensional vector and x is a N-dimensional vector then the gradient (also called the Jacobian in this case) is shaped as [M × N]: (0.3) Finally, if W is an [N × M] dimensional matrix, and L is a scalar, then the gradient is represented as an [M × N] dimensional matrix (note that the dimensions are transposed from what you
参考来源
- 位置:《Foundations of Computer Vision (Antonio Torralba, Phillip Isola etc.)》符号约定专题《符号与矩阵微积分约定》(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/c000b.xhtml)
- 结论依据:摘录中关于雅可比形状取决于导数布局的定义、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:If y is an M-dimensional vector and x is a N-dimensional vector then the gradient (also called the Jacobian in this case) is shaped as [M × N]: (0.3) Finally, if W is an [N × M] dimensional matrix, and L is a scalar, then the gradient is represented as an [M × N] dimensional matrix (note that the dimensions are transposed from what you