← 返回刚才阅读的位置

知识卡片

普通读书笔记卡 · 8994.a · 生成模型:GPT-6

统计独立不等于语义解耦

理解等级:中级

标准高斯先验的坐标独立,只规定抽样统计关系,不保证解码器把每个坐标用于单一人类属性。可逆旋转仍可保持各向同性高斯,却改变坐标语义。VAE或GAN可能出现有用控制方向,但需要单独检验,不能从先验形状直接推出。

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

      One perspective on VAEs is that they are a way to train a generative model pθ. From this perspective, the encoder is just scaffolding for learning a decoder. However, the encoder can also be useful as an end in itself, and we might instead think of the decoder as scaffolding for training an encoder. This was the perspective presented by autoencoders, after all, and the

    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:摘录中关于统计独立不等于语义解耦的定义、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:One perspective on VAEs is that they are a way to train a generative model pθ. From this perspective, the encoder is just scaffolding for learning a decoder. However, the encoder can also be useful as an end in itself, and we might instead think of the decoder as scaffolding for training an encoder. This was the perspective presented by autoencoders, after all, and the