知识卡片

普通读书笔记卡 · 8991.d · 生成模型:GPT-6

简单条件不保证简单后验

理解等级:中级

即使先验p(z)为高斯、给定z的观测pθ(x|z)也为高斯,非线性解码映射仍可令pθ(z|x)复杂或多峰。不能根据生成两端的简单分布,就认为后验必为高斯或只有一个可恢复的潜状态。

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

      Remember, we have defined simple forms for only pθ(X | z) and pz (both are Gaussians), but this does not mean pθ(Z | x) has a simple form. Sampling from arbitrary distributions is a standard topic in statistics, and many algorithms have been proposed, including the Markov chain Monte Carlo methods we encountered in previous chapters. VAEs use a strategy called variational inference. The name

    参考来源

    - 位置:《Foundations of Computer Vision (Antonio Torralba, Phillip Isola etc.)》第33章《生成建模与表示学习》(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/c033.xhtml) - 结论依据:摘录中关于简单条件不保证简单后验的定义、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:Remember, we have defined simple forms for only pθ(X | z) and pz (both are Gaussians), but this does not mean pθ(Z | x) has a simple form. Sampling from arbitrary distributions is a standard topic in statistics, and many algorithms have been proposed, including the Markov chain Monte Carlo methods we encountered in previous chapters. VAEs use a strategy called variational inference. The name