Skip to content

Commit

Permalink
slightly adjust
Browse files Browse the repository at this point in the history
Former-commit-id: 1657ed13906431c9cde3ec4f9e25d0c5dc40a8ba
  • Loading branch information
futianfan committed Jan 31, 2017
1 parent be9db1c commit a29f384
Showing 1 changed file with 6 additions and 2 deletions.
8 changes: 6 additions & 2 deletions Chapter19/approximate_inference.tex
Original file line number Diff line number Diff line change
Expand Up @@ -29,6 +29,8 @@ \chapter{\glsentrytext{approximate_inference}}
% 623 end



% 624 head
\begin{figure}[!htb]
\ifOpenSource
\centerline{\includegraphics{figure.pdf}}
Expand All @@ -46,6 +48,8 @@ \chapter{\glsentrytext{approximate_inference}}
举个例子,\gls{PPCA}的图结构如右图所示,然而由于其条件分布的特殊性质(带有相互正交基向量的线性高斯条件分布)依然能够进行简单的推断。}
\label{fig:intractable_graphs}
\end{figure}
% 624 head



\section{推断是一个优化问题}
Expand All @@ -64,7 +68,7 @@ \section{推断是一个优化问题}
这个下界叫做\firstall{ELBO}。
这个下界的另一个常用名字是负\firstgls{variational_free_energy}。
具体地,这个\gls{ELBO}是这样定义的:

% 625 head
\begin{align}
\CalL(\Vv,{\Vtheta},q) = \log p(\Vv;{\Vtheta}) - D_{\text{KL}}(q(\Vh\mid\Vv) \Vert p(\Vh\mid\Vv;{\Vtheta})).
\end{align}
Expand All @@ -78,7 +82,7 @@ \section{推断是一个优化问题}
% 625


令人吃惊的是,对某些分布$q$$\CalL$可以被化得更简单
令人吃惊的是,对某些分布$q$计算$\CalL$可以变得更简单
通过简单的代数运算我们可以把$\CalL$重写成一个更加简单的形式:

\begin{align}
Expand Down

0 comments on commit a29f384

Please sign in to comment.