← 返回刚才阅读的位置

知识卡片

普通读书笔记卡 · 8996.d · 生成模型:GPT-6

平方损失不证明真实输出独立

理解等级:中级

平方误差可解释为固定对角协方差高斯观测模型的负对数似然,但使用该损失本身只规定均值估计的风险,不证明真实目标各维条件独立。预测函数也可联合使用全部输入。书中从损失分解推出真实独立的表述应限定为所选择的概率模型假设。

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 underlying assumption of this factorization is one of conditional independence: each pixel's ab value is considered to be conditionally independent from all other pixels’ ab values, given the observed luminance (of all pixels). This is a very common assumption in image modeling problems: in fact, whenever you use least-squares regression for a multidimensional prediction, you are implicitly making this same assumption. To see this,

    参考来源

    - 位置:《Foundations of Computer Vision (Antonio Torralba, Phillip Isola etc.)》第34章《条件生成模型》(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/c034.xhtml) - 结论依据:摘录中关于平方损失不证明真实输出独立的定义、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 underlying assumption of this factorization is one of conditional independence: each pixel's ab value is considered to be conditionally independent from all other pixels’ ab values, given the observed luminance (of all pixels). This is a very common assumption in image modeling problems: in fact, whenever you use least-squares regression for a multidimensional prediction, you are implicitly making this same assumption. To see this,