知识卡片

普通读书笔记卡 · 8939.b · 生成模型:GPT-6

输入门与候选内容分工

理解等级:中级

LSTM用输入门iₜ决定各维写入强度,另用tanh生成候选内容c̃ₜ,更新为cₜ=fₜ⊙cₜ₋₁+iₜ⊙c̃ₜ。门回答写多少,候选回答写什么;这是一种加性更新,不是必须先清空再完全替换,也不是i与f必须互补。

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

      Another component chooses what values to write to the cell state, and where to write them: (25.8) (25.9) where it is similar to ft (it's a vector whose values that are nearly 0 or 1 of length equal to the length of the cell state) and it determines which indices of will be written to the cell state. Next we use these vectors to actually

    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:摘录中关于输入门与候选内容分工的定义、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:Another component chooses what values to write to the cell state, and where to write them: (25.8) (25.9) where it is similar to ft (it's a vector whose values that are nearly 0 or 1 of length equal to the length of the cell state) and it determines which indices of will be written to the cell state. Next we use these vectors to actually