知识卡片

普通读书笔记卡 · 8926.c · 生成模型:GPT-6

卷积带偏置是仿射层

理解等级:中级

无偏置的卷积可表示为具有共享结构的线性矩阵,加入偏置后是仿射映射。二维有限图像常对应块Toeplitz或其他由边界决定的结构,周期边界则形成循环结构。参数共享让矩阵中的不同连接共用核权重,减少独立参数。

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

      As discussed in chapter 15, convolution is just a special kind of linear transform. Similarly, a convolutional layer is just a special kind of linear layer. It is a linear layer whose matrix W is Toeplitz. We can view it either as a matrix or as a neural net, as shown in figure 24.4, which shows the case of a one-dimensional (1D) convolution over a

    参考来源

    - 位置:《Foundations of Computer Vision (Antonio Torralba, Phillip Isola etc.)》第24章《卷积神经网络》(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/c024b.xhtml) - 结论依据:摘录中关于卷积带偏置是仿射层的定义、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:As discussed in chapter 15, convolution is just a special kind of linear transform. Similarly, a convolutional layer is just a special kind of linear layer. It is a linear layer whose matrix W is Toeplitz. We can view it either as a matrix or as a neural net, as shown in figure 24.4, which shows the case of a one-dimensional (1D) convolution over a