← 返回刚才阅读的位置

知识卡片

普通读书笔记卡 · 9169.b · 生成模型:GPT-6

平滑激活让梯度携带方向信息

理解等级:中级

Sigmoid把输入映到0与1之间,中心附近导数较大,两端趋于饱和;tanh则以零为中心。平滑便于链式求导,但多层中的小导数仍会使梯度衰减;数值接近0不等于神经元数学上永不激活。

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

      Using an on-off Heaviside step function makes training a neural network using gradient descent rather difficult, as a function that switches from “not working at all” to “working completely” provides very little information concerning in which direction to move. It is therefore more desirable to have a smoother activation function. One such function is the sigmoid function:

    参考来源

    - 位置:《Introduction to Autonomous Robots Mechanisms, Sensors, Actuators, and Algorithms (Nikolaus Correll, Bradley Hayes etc.)》第10章《人工神经网络》(来源项目:a19c7933-4eda-4176-aebd-30fe7e1ad292;源文件:extracted/Introduction to Autonomous Robots Mechanisms, Sensors, Actuators, and Algorithms (Nikolaus Correll, Bradley Hayes etc.)/_epub-src/OEBPS/xhtml/chapter_10.xhtml) - 结论依据:摘录中关于平滑激活让梯度携带方向信息的定义、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:Using an on-off Heaviside step function makes training a neural network using gradient descent rather difficult, as a function that switches from “not working at all” to “working completely” provides very little information concerning in which direction to move. It is therefore more desirable to have a smoother activation function. One such function is the sigmoid function: