← 返回刚才阅读的位置

知识卡片

普通读书笔记卡 · 8994.c · 生成模型:GPT-6

发现的因素未必是真实机制

理解等级:中级

河流示例中VAE得到控制整体弯曲的方向,但原程序用许多局部硬币转向而非一个全局曲率参数。不同潜机制可能解释同一观测分布,所以可解释控制变量不等于恢复真实生成原因。要作因果识别,还需额外数据或结构假设。

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

      One of the latent dimensions seems to control the grass color, and another controls the river curvature! These two latent dimensions are disentangled in the sense that varying the latent dimension that controls color has little effect on curvature and varying the latent dimension that controls curvature has little effect on color. Indeed, grass color was one of the attributes of the true data generating

    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:摘录中关于发现的因素未必是真实机制的定义、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:One of the latent dimensions seems to control the grass color, and another controls the river curvature! These two latent dimensions are disentangled in the sense that varying the latent dimension that controls color has little effect on curvature and varying the latent dimension that controls curvature has little effect on color. Indeed, grass color was one of the attributes of the true data generating