← 返回刚才阅读的位置

知识卡片

普通读书笔记卡 · 8872.c · 生成模型:GPT-6

前向激活可参数化反向网络

理解等级:高级

ReLU前向输入决定反向门控矩阵,原权重决定反向转置矩阵,因此可将前向过程看作生成反向网络参数的超网络。这个视角解释前向信息为何要传给反向模块,不表示反向中存在一套独立训练的新权重。

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

      things about this forward-backward network. One is that activations from the relu layers get transformed to become parameters of a linear layer of the backward network (see equation (14.33)). There is a general term for this setup, where one neural net outputs values that parameterize another neural net; this is called a hypernetwork [179]. The forward network is a hypernetwork that parameterizes the backward network.

    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:摘录中关于前向激活可参数化反向网络的定义、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:things about this forward-backward network. One is that activations from the relu layers get transformed to become parameters of a linear layer of the backward network (see equation (14.33)). There is a general term for this setup, where one neural net outputs values that parameterize another neural net; this is called a hypernetwork [179]. The forward network is a hypernetwork that parameterizes the backward network.