← 返回刚才阅读的位置

知识卡片

普通读书笔记卡 · 6724 · 生成模型:GPT-6

多核宽分支可以并行汇聚

理解等级:中级

不同核宽各自卷积、激活及池化后拼接,捕捉多种局部范围,不是把所有卷积层逐层堆叠。若5种核宽各5输出通道并全局池化,每样本得25个值;输出宽度不必等嵌入维数。

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

      Unlike in the previous examples, this time you’ll create multiple convolution and pooling layers. You won’t layer them up, as is often done in computer vision; instead, you will concatenate the convolution and pooling outputs together. This is effective because you’ve limited the dimensionality of your convolution and pooling output by performing global max pooling and keeping the number of output channels much smaller than

    参考来源

    - 位置:《Natural Language Processing in Action, Second Edition》第07章《卷积网络识别文本模式》(来源项目:a19c7933-4eda-4176-aebd-30fe7e1ad292;源文件:extracted/Natural Language Processing in Action, Second Edition/_epub-src/OEBPS/Text/chapter-7.html) - 结论依据:摘录中关于多核宽分支可以并行汇聚的定义、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:Unlike in the previous examples, this time you’ll create multiple convolution and pooling layers. You won’t layer them up, as is often done in computer vision; instead, you will concatenate the convolution and pooling outputs together. This is effective because you’ve limited the dimensionality of your convolution and pooling output by performing global max pooling and keeping the number of output channels much smaller than