知识卡片
重建与潜约束需要平衡
重建倾向保留可区分输入的信息,KL倾向让每个近似后验接近先验,二者存在张力。若编码器输出与输入几乎无关的先验,潜变量可能不再传递有用信息。VAE并不保证自动形成无空洞、充分利用且良好重建的潜空间。
这张卡片看不懂?
这张卡片已通过人工审查,不再接受失败反馈。
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 KL term encourages the encodings z(i) to be near a unit Gaussian distribution. To get an intuition for the effect of this term, consider the case where we fix to be 1; this is still a valid model for qψ(Z|x), just with lower capacity because it has fewer free parameters. In this simple case, the KL term reduces to . The effect of this
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:摘录中关于重建与潜约束需要平衡的定义、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 KL term encourages the encodings z(i) to be near a unit Gaussian distribution. To get an intuition for the effect of this term, consider the case where we fix to be 1; this is still a valid model for qψ(Z|x), just with lower capacity because it has fewer free parameters. In this simple case, the KL term reduces to . The effect of this