← 返回刚才阅读的位置

知识卡片

普通读书笔记卡 · 6686.f · 生成模型:GPT-6

硬阈值与平滑激活用途不同

理解等级:中级

硬阈值把分数变成0或1,会丢失幅度,普通梯度又几乎处处为0。训练可采用适合求导的激活或代理方法,推断再按决策阈值分类;硬阈值并非数学上绝对只能放输出层,但不宜直接当普通可微隐藏单元。

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

      There’s one more piece you’re missing here. You need to run a nonlinear function on the output (y) to change the shape of the output, so it’s not just a linear regression. Often, a thresholding or clipping function is used to decide whether or not the neuron should fire. For a thresholding function, if the weighted sum is above a certain threshold, the perceptron outputs

    参考来源

    - 位置:《Natural Language Processing in Action, Second Edition》第05章《神经网络基础》(来源项目:a19c7933-4eda-4176-aebd-30fe7e1ad292;源文件:extracted/Natural Language Processing in Action, Second Edition/_epub-src/OEBPS/Text/chapter-5.html) - 结论依据:摘录中关于硬阈值与平滑激活用途不同的定义、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:There’s one more piece you’re missing here. You need to run a nonlinear function on the output (y) to change the shape of the output, so it’s not just a linear regression. Often, a thresholding or clipping function is used to decide whether or not the neuron should fire. For a thresholding function, if the weighted sum is above a certain threshold, the perceptron outputs