← 返回起始笔记位置

知识卡片

普通读书笔记卡 · 8929.a · 生成模型:GPT-6

可分离核降低空间计算

理解等级:中级

单通道K×K的秩一核可写uvᵀ,先做K×1再做1×K滤波即可,计算由每点K²乘加降为约2K。中间若插入非线性,就不再等价于同一个线性可分离核;多中间通道则能表示多个外积之和,不能一概称秩一。

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

      Separable filters are another kind of low-rank filter that is useful in many applications (see chapter 16). We can create a convolutional layer with separable filters by simply stacking two convolutional layers in sequence, with no other layers in between. The first layer is a filter bank with K × 1 kernels and the second uses 1 × K kernels. The composition of these layers

    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:摘录中关于可分离核降低空间计算的定义、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:Separable filters are another kind of low-rank filter that is useful in many applications (see chapter 16). We can create a convolutional layer with separable filters by simply stacking two convolutional layers in sequence, with no other layers in between. The first layer is a filter bank with K × 1 kernels and the second uses 1 × K kernels. The composition of these layers