← 返回起始笔记位置

知识卡片

普通读书笔记卡 · 8930.a · 生成模型:GPT-6

局部池化只有限度容忍位移

理解等级:中级

只要特征在同一汇聚窗口内移动且不改变最大值,max池化输出可以保持不变。但跨越窗口边界或抽取相位变化时,输出仍可能突变。局部池化提供的是受窗口限制的容忍,不保证任意平移不变;全局汇聚也要考虑边界是否使特征丢失。

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

      Like all downsampling layers, pooling layers can be used to reduce the resolution of the input tensor, removing high-frequency information in the signal. Pooling is also particularly useful as a way to achieve invariance. Convolutional layers produce outputs that are equivariant to translations of their input. Pooling is a way to convert equivariance into invariance. For example, suppose we have run a convolutional filter that

    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:摘录中关于局部池化只有限度容忍位移的定义、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:Like all downsampling layers, pooling layers can be used to reduce the resolution of the input tensor, removing high-frequency information in the signal. Pooling is also particularly useful as a way to achieve invariance. Convolutional layers produce outputs that are equivariant to translations of their input. Pooling is a way to convert equivariance into invariance. For example, suppose we have run a convolutional filter that