← 返回刚才阅读的位置

知识卡片

普通读书笔记卡 · 9169.c · 生成模型:GPT-6

ReLU保留正区梯度

理解等级:中级

ReLU=max(0,x)在正侧导数为1、负侧为0,泄漏ReLU为负侧保留小斜率。它计算简单且改变梯度传播,但零点需选次梯度约定,负区也可能长期无更新;0.1只是泄漏斜率例子,不是固定定义。

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

      and is shown on the right side of figure 10.3. The dashed line indicates a refinement of the ReLU known as leaky ReLU, with a typical slope of 0.1; it improves learning for negative wx by providing a directional gradient.

    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ReLU保留正区梯度的定义、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:and is shown on the right side of figure 10.3. The dashed line indicates a refinement of the ReLU known as leaky ReLU, with a typical slope of 0.1; it improves learning for negative wx by providing a directional gradient.