知识卡片
局部雅可比与目标梯度不同
局部雅可比描述模块输出对模块输入或参数的敏感度,上游梯度描述最终损失对模块输出的敏感度。二者都在同一次前向的当前值处求值,乘起来才得到最终目标对模块输入的导数。混用不同参数版本会破坏链式法则对应关系。
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
- · 反向传播
原书出处 · 1 处
- · 查看原书依据
All these arrays represent the gradient at a single operating point, namely that of the current value of the data and parameters. (14.4) (14.5) (14.6) (14.7) These arrays give a simple formula for computing the gradient we need, that is, , in order to update θ to minimize the cost: (14.8) (14.9) The transpose is because, by convention, θ is a column vector while is
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:摘录中关于局部雅可比与目标梯度不同的定义、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:All these arrays represent the gradient at a single operating point, namely that of the current value of the data and parameters. (14.4) (14.5) (14.6) (14.7) These arrays give a simple formula for computing the gradient we need, that is, , in order to update θ to minimize the cost: (14.8) (14.9) The transpose is because, by convention, θ is a column vector while is