知识卡片

普通读书笔记卡 · 8878.c · 生成模型:GPT-6

固定核方向与学习核区别对待

理解等级:中级

网络常把实际相关运算称作卷积。若核自由学习,中心反转可通过重新参数化表示相同映射集合;若核已由物理模型或方向导数固定,换算子会改变语义。相关通常不具卷积同样的交换结合律,不能仅依名称混用代数性质。

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

      Most of the literature in neural networks calls convolution to the correlation operator. As the kernels are learned, the distinction is not important most of the time. One important property of the convolution is that it is commutative: h [n] ◦ ℓ [n] = ℓ [n] ◦ h [n]. This property is easy to prove by changing the variables, k = n – k′, so

    参考来源

    - 位置:《Foundations of Computer Vision (Antonio Torralba, Phillip Isola etc.)》第15章《线性图像滤波》(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/c015b.xhtml) - 结论依据:摘录中关于固定核方向与学习核区别对待的定义、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:Most of the literature in neural networks calls convolution to the correlation operator. As the kernels are learned, the distinction is not important most of the time. One important property of the convolution is that it is commutative: h [n] ◦ ℓ [n] = ℓ [n] ◦ h [n]. This property is easy to prove by changing the variables, k = n – k′, so