Skip to content

Commit

Permalink
modify chap 20, fix exacity#148, fix exacity#149
Browse files Browse the repository at this point in the history
  • Loading branch information
SwordYork committed Sep 28, 2017
1 parent 54eb6c4 commit f462551
Showing 1 changed file with 2 additions and 3 deletions.
5 changes: 2 additions & 3 deletions Chapter20/deep_generative_models.tex
Original file line number Diff line number Diff line change
Expand Up @@ -388,8 +388,7 @@ \subsection{有趣的性质}


\glssymbol{DBM}~一个不理想的特性是从中采样是相对困难的。
\glssymbol{DBN}~只需要在其顶部的一对层中使用~\glssymbol{mcmc}~采样。
其他层仅在采样过程末尾涉及,并且只需在一个高效的\gls{ancestral_sampling}过程。
在一次高效的\gls{ancestral_sampling}过程中,\glssymbol{DBN}~只需要在其顶部的一对层中使用~\glssymbol{mcmc}~采样,而其他层仅在采样过程末尾参与。
要从~\glssymbol{DBM}~生成样本,必须在所有层中使用~\glssymbol{mcmc},并且模型的每一层都参与每个\gls{markov_chain}转移。


Expand Down Expand Up @@ -455,7 +454,7 @@ \subsection{\glssymbol{DBM}\glsentrytext{meanfield}\gls{inference}}

应用这些一般的方程,我们得到以下更新规则(再次忽略\gls{bias_aff}项):
\begin{align} \label{eq:2033h1}
h_j^{(1)} &= \sigma \Big( \sum_i v_i \MW_{i,j}^{(1)}
\hat h_j^{(1)} &= \sigma \Big( \sum_i v_i \MW_{i,j}^{(1)}
+ \sum_{k^{\prime}} \MW_{j,k^{\prime}}^{(2)} \hat h_{k^{\prime}}^{(2)} \Big), ~\forall j ,\\
\label{eq:2034h2}
\hat h_{k}^{(2)} &= \sigma \Big( \sum_{j^{\prime}} \MW_{j^{\prime},k}^{(2)}
Expand Down

0 comments on commit f462551

Please sign in to comment.