← 返回起始笔记位置

知识卡片

普通读书笔记卡 · 8989.a · 生成模型:GPT-6

反向传播不是函数反演

理解等级:初级

反向传播沿计算图传播导数,用于计算损失对输入或参数的梯度;反演则寻找能产生某个输出的输入。即使二者都在图上向后处理,计算目标也不同。自动求导不会自动给出生成器的逆或观测对应的唯一潜变量。

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

      Moving backward through a net is also what backpropagation does, but it computes a different function: backpropagation computes the gradient ∇f whereas here we focus on the inverse f–1. 33.2 Latent Variables as Representations In chapter 32, we introduced generative models with latent variables z. In that context, the role of the latent variable was to specify all unobserved factors that might affect the output

    参考来源

    - 位置:《Foundations of Computer Vision (Antonio Torralba, Phillip Isola etc.)》第33章《生成建模与表示学习》(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/c033.xhtml) - 结论依据:摘录中关于反向传播不是函数反演的定义、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:Moving backward through a net is also what backpropagation does, but it computes a different function: backpropagation computes the gradient ∇f whereas here we focus on the inverse f–1. 33.2 Latent Variables as Representations In chapter 32, we introduced generative models with latent variables z. In that context, the role of the latent variable was to specify all unobserved factors that might affect the output