← 返回刚才阅读的位置

知识卡片

普通读书笔记卡 · 8938.b · 生成模型:GPT-6

堆叠深度与时间长度不同

理解等级:中级

深层RNN在同一时间步把较低层状态送到较高层,每层又沿时间维保持自己的递归状态。增加层数扩展每步处理深度,增加序列长度扩展递归次数,两者对应不同依赖路径。代码实现须为每层保留独立状态,并共享各层跨时间的权重。

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

      A deep RNN can be constructed by stacking recurrent layers. Here is an example: Figure 25.6: A deep RNN with multiple hidden layers. To emphasize that the parameters are shared, we only show them once. Here is code for a two-layer version of this RNN: We set the bias to be “False” for U1 and U2 since the recurrent layer already gets a bias term

    参考来源

    - 位置:《Foundations of Computer Vision (Antonio Torralba, Phillip Isola etc.)》第25章《循环神经网络》(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/c025.xhtml) - 结论依据:摘录中关于堆叠深度与时间长度不同的定义、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:A deep RNN can be constructed by stacking recurrent layers. Here is an example: Figure 25.6: A deep RNN with multiple hidden layers. To emphasize that the parameters are shared, we only show them once. Here is code for a two-layer version of this RNN: We set the bias to be “False” for U1 and U2 since the recurrent layer already gets a bias term