← 返回刚才阅读的位置

知识卡片

普通读书笔记卡 · 8997.f · 生成模型:GPT-6

条件变分编码器训练看见目标

理解等级:中级

训练cVAE时,用qψ(z|x,y)近似给定条件与目标的后验,解码器读取x和z来解释y。生成新y时目标尚未知,应从所定义的p(z|x)或独立先验抽样。不能把训练时利用目标的编码路径误带进推理,造成答案泄漏。

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

      Just like regular VAEs, cVAEs also have an encoder, which acts to predict the optimal importance sampling distribution pθ(Z | x, y). In practice, this means that a cVAE can be trained just like a regular VAE except that the encoder takes the conditioning information, x, as input (in addition to y), and the decoder also takes in x (in addition to z). Figure 34.5

    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:摘录中关于条件变分编码器训练看见目标的定义、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:Just like regular VAEs, cVAEs also have an encoder, which acts to predict the optimal importance sampling distribution pθ(Z | x, y). In practice, this means that a cVAE can be trained just like a regular VAE except that the encoder takes the conditioning information, x, as input (in addition to y), and the decoder also takes in x (in addition to z). Figure 34.5