知识卡片

普通读书笔记卡 · 8997.a · 生成模型:GPT-6

共享潜变量协调多个输出

理解等级:中级

结构化条件生成可以通过共享潜状态、自回归依赖或能量关系,让多个输出位置作相容选择。例如一次为整件衣服选择某种颜色,再生成相应像素。模型设计仍需检验是否实际使用共同随机性,不能只因名为条件生成就保证多峰与结构都学好。

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

      Methods that model joint structure in the output space are called structured prediction methods—they don't factorize the output into independent potentials conditioned on the input. Conditional generative modeling is a structured prediction approach that models a full distribution of possibilities over the joint configuration of outputs.

    参考来源

    - 位置:《Foundations of Computer Vision (Antonio Torralba, Phillip Isola etc.)》第34章《条件生成模型》(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/c034.xhtml) - 结论依据:摘录中关于共享潜变量协调多个输出的定义、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:Methods that model joint structure in the output space are called structured prediction methods—they don't factorize the output into independent potentials conditioned on the input. Conditional generative modeling is a structured prediction approach that models a full distribution of possibilities over the joint configuration of outputs.