知识卡片
反向传播复用链式法则公共项
不同层参数的梯度共享从后续层到损失的导数乘积。反向传播从输出向前复用这些公共项,避免每个参数独立重算整条链。这是对链式法则计算过程的组织,不是另一个近似梯度公式,也不是把前向函数求逆。
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
then can use it to compute both and . Now notice that this pattern of reuse can be applied in the same way for θ3, θ4, and so on. This is the whole trick of backpropagation: rather than computing each layer's gradients independently, observe that they share many of the same terms, so we might as well calculate each shared term once and reuse them.
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:摘录中关于反向传播复用链式法则公共项的定义、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:then can use it to compute both and . Now notice that this pattern of reuse can be applied in the same way for θ3, θ4, and so on. This is the whole trick of backpropagation: rather than computing each layer's gradients independently, observe that they share many of the same terms, so we might as well calculate each shared term once and reuse them.