← 返回起始笔记位置

知识卡片

普通读书笔记卡 · 8868.a · 生成模型:GPT-6

逐层表示也有各自的分布

理解等级:中级

固定网络对输入分布逐层作用,每个隐藏层都产生新的激活分布。表示变换不仅改变单一样本坐标,也会改变样本密度与几何形状。这里的分布是数据通过映射形成的结果,不等于每个样本的类别概率向量,也不要求层具有显式密度模型。

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 distribution of activations on some layer ℓ to be pℓ. Then, layer by layer, a deep net transforms pdata into p1 into p2, and so on until finally transforming the data to the distribution pout. Most loss functions can also be interpreted from this angle: they penalize the divergence, in one form or another, between the output distribution pout and a target distribution ptarget.

    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:摘录中关于逐层表示也有各自的分布的定义、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 distribution of activations on some layer ℓ to be pℓ. Then, layer by layer, a deep net transforms pdata into p1 into p2, and so on until finally transforming the data to the distribution pout. Most loss functions can also be interpreted from this angle: they penalize the divergence, in one form or another, between the output distribution pout and a target distribution ptarget.