知识卡片
遗忘门控制旧状态保留率
遗忘门fₜ由当前输入和上一隐藏态经Sigmoid生成,逐元素乘旧细胞状态。接近1表示多保留,接近0表示强衰减。它是连续软门,不必接近二值,也不是直接“选索引删除”;有限参数下Sigmoid通常不精确达到0或1。
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
- · 循环神经网络
原书出处 · 1 处
- · 查看原书依据
One component picks the indices of the cell state to delete: (25.7) where ft is a vector of length equal to the length of the cell state ct, and σ is the sigmoid function that outputs values in the range [0, 1]. The idea is that most values in ft will be near either 1 (remember) or 0 (forget). Later ft will be pointwise multiplied
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:摘录中关于遗忘门控制旧状态保留率的定义、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:One component picks the indices of the cell state to delete: (25.7) where ft is a vector of length equal to the length of the cell state ct, and σ is the sigmoid function that outputs values in the range [0, 1]. The idea is that most values in ft will be near either 1 (remember) or 0 (forget). Later ft will be pointwise multiplied