知识卡片

普通读书笔记卡 · 8866.c · 生成模型:GPT-6

批归一化先标准化再仿射

理解等级:中级

批归一化按指定通道的批统计量标准化激活,再使用可学习γ和β缩放平移,保留非零均值与非单位尺度的表达能力。标准化轴必须明确,实际计算还应处理零方差等数值情况,不能把省略稳定项的教学公式直接当完整实现。

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

      Batchnorm standardizes each neural activation with respect to its mean and variance over a batch of datapoints. Mathematically, (12.23) Recall from statistics that the standard score of a draw of a random variable is how many standard deviations it differs from the mean: z = . where γ and β are learned parameters of this layer that maintain expressivity so that the layer can output

    参考来源

    - 位置:《Foundations of Computer Vision (Antonio Torralba, Phillip Isola etc.)》第12章《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/c012.xhtml) - 结论依据:摘录中关于批归一化先标准化再仿射的定义、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:Batchnorm standardizes each neural activation with respect to its mean and variance over a batch of datapoints. Mathematically, (12.23) Recall from statistics that the standard score of a draw of a random variable is how many standard deviations it differs from the mean: z = . where γ and β are learned parameters of this layer that maintain expressivity so that the layer can output