知识卡片
改变池化范围需要重算输出尺寸
核覆盖整个卷积响应时,每通道仅保留1项;局部池化则仍可有多个位置。更改卷积或池化参数后,应逐层重算输出长度及拼接总宽,再设置线性层,不能照抄另一版本1856维或25维。
这张卡片看不懂?
这张卡片已通过人工审查,不再接受失败反馈。
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
Next, you can build the convolution and pooling layers. Using the output size of each convolution layer, you can define a pooling layer, whose kernel takes up the entire convolutional layer output sequence, as shown in the following listing. In fact, this is the same strategy used by NLP experts like Christopher Manning26 and Yoon Kim27 to achieve global max pooling and state-of-the-art performance by
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:摘录中关于改变池化范围需要重算输出尺寸的定义、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:Next, you can build the convolution and pooling layers. Using the output size of each convolution layer, you can define a pooling layer, whose kernel takes up the entire convolutional layer output sequence, as shown in the following listing. In fact, this is the same strategy used by NLP experts like Christopher Manning26 and Yoon Kim27 to achieve global max pooling and state-of-the-art performance by