← 返回起始笔记位置

知识卡片

普通读书笔记卡 · 8984.c · 生成模型:GPT-6

高斯归一常数随方差变化

理解等级:中级

一维高斯以均值μ和方差v>0参数化,p(x)=(2πv)⁻¹ᐟ² exp(−(x−μ)²/(2v))。若写成exp项除Z,则Z=√(2πv),不是其倒数。优化方差时必须包含log Z对v的依赖,否则目标会变错。

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

      This density has two parameters θ1 and θ2, which are the mean and variance of the distribution. The normalization constant Z ensures that the function is normalized. This is the typical strategy in defining density models: create a parameterized family of functions such that any function in the family is normalized. Given such a family, we search over the parameters to optimize a generative modeling

    参考来源

    - 位置:《Foundations of Computer Vision (Antonio Torralba, Phillip Isola etc.)》第32章《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/c032.xhtml) - 结论依据:摘录中关于高斯归一常数随方差变化的定义、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:This density has two parameters θ1 and θ2, which are the mean and variance of the distribution. The normalization constant Z ensures that the function is normalized. This is the typical strategy in defining density models: create a parameterized family of functions such that any function in the family is normalized. Given such a family, we search over the parameters to optimize a generative modeling