知识卡片

普通读书笔记卡 · 8994.b · 生成模型:GPT-6

潜变量遍历检验局部控制

理解等级:中级

固定其他潜坐标,沿某一方向改变取值并观察输出,可发现颜色、曲率或姿态等变化方向。结果只覆盖选定起点和范围,未变化属性也需明确观察;少量网格可视化不足以证明全空间都独立控制同一属性。

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

      Did the VAE uncover the true latent variables that generated the data, that is, did it recover latent dimensions corresponding to the attribute values that were the inputs to the Python script? We can examine this by generating a set of images that walk along two latent dimensions of the VAE's z-space, shown in figure 33.14.

    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:摘录中关于潜变量遍历检验局部控制的定义、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:Did the VAE uncover the true latent variables that generated the data, that is, did it recover latent dimensions corresponding to the attribute values that were the inputs to the Python script? We can examine this by generating a set of images that walk along two latent dimensions of the VAE's z-space, shown in figure 33.14.