知识卡片

结构图卡 · 8939.e · 生成模型:GPT-6

门控状态更新的路线

理解等级:中级

结构图:

flowchart LR
 A[当前输入与上一隐藏态] --> B[遗忘门f]
 A --> C[输入门i与候选内容]
 D[上一细胞状态] --> E[保留旧值加写入候选]
 B --> E
 C --> E
 E --> F[新细胞状态]
 A --> G[输出门o]
 F --> H[tanh后按输出门读出]
 G --> H
 H --> I[新隐藏态]

这些软门决定状态如何流动,不能由名称推断模型在实际任务中已经学会正确记忆。

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

      First, the cell state is updated based on the incoming signals, ht–1 and xin[t]. Then the cell state is used to determine the hidden state ht to output. Different subcomponents of the LSTM layer decide what to delete from the cell state, what to write to the cell state, and what to read off the cell state.

    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:摘录中关于门控状态更新的路线的定义、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:First, the cell state is updated based on the incoming signals, ht–1 and xin[t]. Then the cell state is used to determine the hidden state ht to output. Different subcomponents of the LSTM layer decide what to delete from the cell state, what to write to the cell state, and what to read off the cell state.