知识卡片

普通读书笔记卡 · 8992 · 生成模型:GPT-6

变分编码器输出分布参数

理解等级:中级

VAE用qψ(z|x)近似模型后验,编码器fψ(x)输出该分布的参数,例如均值和方差。编码过程包含“确定参数”和“随机抽样”两步,输出不只是固定向量。后验近似受分布族容量与共享编码网络能力限制。

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

      The idea of variational inference is to approximate an intractable density p by finding the nearest density in a tractable family qψ, parameterized by ψ. In VAEs, we approximate our ideal importance density pθ(Z | x) with a qψ in a family we can efficiently sample from; the most common choice is to again use a Gaussian, conditioned on x: . The fψ is a

    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:摘录中关于变分编码器输出分布参数的定义、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:The idea of variational inference is to approximate an intractable density p by finding the nearest density in a tractable family qψ, parameterized by ψ. In VAEs, we approximate our ideal importance density pθ(Z | x) with a qψ in a family we can efficiently sample from; the most common choice is to again use a Gaussian, conditioned on x: . The fψ is a