知识卡片

普通读书笔记卡 · 8984.b · 生成模型:GPT-6

采样能否可控要单独评估

理解等级:中级

显式潜变量常便于对同一随机输入进行插值或编辑;密度模型也可通过采样算法的随机性生成不同样本。是否有低维、可解释、易反演的控制空间,与是否能抽样是不同能力。书中模型表是典型用途概览,不是严格的不可能性结论。

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

      Each of these families of model has its own advantages and disadvantages. Generators have latent variables z that control the properties of the generated image. Changing these variables can change the generated image in meaningful ways; we will explore this idea in greater detail in the next chapter. In contrast, density and energy models do not have latent variables that directly control generated samples.

    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:摘录中关于采样能否可控要单独评估的定义、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:Each of these families of model has its own advantages and disadvantages. Generators have latent variables z that control the properties of the generated image. Changing these variables can change the generated image in meaningful ways; we will explore this idea in greater detail in the next chapter. In contrast, density and energy models do not have latent variables that directly control generated samples.