知识卡片
高斯先验的散度可解析
若q(z|x)=N(μ,diag(v))、p(z)=N(0,I),则KL=½Σj(μj²+vj−log vj−1),其中vj>0。该式推动均值靠近零、方差靠近1;它约束整个条件分布,不是简单把每个随机潜样本都压到零。
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
For each datapoint x, the KL term can be computed in closed form since it is between two normal distributions (see the appendix of [261] for a derivation): (33.40) (33.41) The other term, the likelihood term, will be approximated by sampling. For each datapoint x, we will take just a single sample from this expectation, as this is often sufficient in practice. To do so,
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:摘录中关于高斯先验的散度可解析的定义、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:For each datapoint x, the KL term can be computed in closed form since it is between two normal distributions (see the appendix of [261] for a derivation): (33.40) (33.41) The other term, the likelihood term, will be approximated by sampling. For each datapoint x, we will take just a single sample from this expectation, as this is often sufficient in practice. To do so,