← 返回起始笔记位置

知识卡片

普通读书笔记卡 · 8868.d · 生成模型:GPT-6

正象限体积比例不是数据概率

理解等级:高级

对按符号对称的空间区域,N维严格正象限占其体积的2⁻ᴺ;这不意味着任意实际激活有同样比例落在其中。数据可能集中在正区,坐标符号也可能相关。几何体积、样本分布与每样本非零坐标数量应分别分析。

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

      the region that is strictly positive occupies only proportion of the embedding space, so almost the entire space gets mapped to the axes after a relu layer. The geometry of high-dimensional neural representations may become very sparse because of this, where most of the volume of representational space is not occupied by any datapoints.

    参考来源

    - 位置:《Foundations of Computer Vision (Antonio Torralba, Phillip Isola etc.)》第13章《作为分布变换器的神经网络》(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/c013.xhtml) - 结论依据:摘录中关于正象限体积比例不是数据概率的定义、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:the region that is strictly positive occupies only proportion of the embedding space, so almost the entire space gets mapped to the axes after a relu layer. The geometry of high-dimensional neural representations may become very sparse because of this, where most of the volume of representational space is not occupied by any datapoints.