知识卡片

普通读书笔记卡 · 8930 · 生成模型:GPT-6

均值池化仍是线性

理解等级:中级

均值池化以固定邻域权重求平均,对输入是线性映射;最大池化选择窗口中的最大值,属于非线性。二者都可配合步长降低分辨率,但max不会像规范低通那样保证删除高频。不能因它们同属池化层就赋予相同频谱性质。

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

      Pooling layers are downsampling layers that summarize the information in a patch using some aggregate statistic, such as the patch's mean value, called mean pooling, or its max value, called max pooling, defined as follows:

    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:摘录中关于均值池化仍是线性的定义、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:Pooling layers are downsampling layers that summarize the information in a patch using some aggregate statistic, such as the patch's mean value, called mean pooling, or its max value, called max pooling, defined as follows: