Skip to content

Commit

Permalink
update ack
Browse files Browse the repository at this point in the history
  • Loading branch information
SwordYork committed May 7, 2017
1 parent b27bdda commit 4f9fca0
Show file tree
Hide file tree
Showing 13 changed files with 542 additions and 21 deletions.
4 changes: 2 additions & 2 deletions Chapter12/applications.tex
Original file line number Diff line number Diff line change
Expand Up @@ -451,10 +451,10 @@ \subsubsection{对比度归一化}
\else
\centering
\begin{tabular}{ccc}
\includegraphics[width=.3\figwidth]{Chapter12/figures/src0.jpg} &
\includegraphics[width=.3\figwidth]{Chapter12/figures/gray0.jpg} &
\includegraphics[width=.3\figwidth]{Chapter12/figures/gcn0.jpg} &
\includegraphics[width=.3\figwidth]{Chapter12/figures/lcn0.jpg} \\
\includegraphics[width=.3\figwidth]{Chapter12/figures/src1.jpg} & % ?? may be problem
\includegraphics[width=.3\figwidth]{Chapter12/figures/gray1.jpg} & % ?? may be problem
\includegraphics[width=.3\figwidth]{Chapter12/figures/gcn1.jpg} &
\includegraphics[width=.3\figwidth]{Chapter12/figures/lcn1.jpg}\\
Input image & GCN & LCN
Expand Down
1 change: 0 additions & 1 deletion Chapter17/monte_carlo_methods.tex
Original file line number Diff line number Diff line change
Expand Up @@ -458,7 +458,6 @@ \section{不同的\glsentrytext{mode}之间的\glsentrytext{mixing}挑战}
但是对于吉布斯链来说从分布的一个\gls{mode}转移到另一个仍然是很困难的,比如说改变数字。
\emph{(右)}从\gls{generative_adversarial_networks}中抽出的连续原始样本。
因为\gls{ancestral_sampling}生成的样本之间互相独立,所以不存在\gls{mixing}问题。
{译者注:原书此处左右搞反了。}}
\label{fig:chap17_fig-dbm-bad-mixing}
\end{figure}
% 593 end
Expand Down
393 changes: 393 additions & 0 deletions Chapter7/annotations.txt

Large diffs are not rendered by default.

2 changes: 1 addition & 1 deletion Chapter7/regularization.tex
Original file line number Diff line number Diff line change
Expand Up @@ -720,7 +720,7 @@ \section{\glsentrytext{early_stopping}}
\ifOpenSource
\centerline{\includegraphics{figure.pdf}}
\else
\centerline{\includegraphics[width=0.8\textwidth]{Chapter7/figures/reg_l1_vs_l2_mistake}}
\centerline{\includegraphics[width=0.8\textwidth]{Chapter7/figures/reg_early_stop_vs_l2}}
\fi
\caption{\gls{early_stopping}效果的示意图。
\emph{(左)}实线轮廓线表示负对数似然的轮廓。
Expand Down
4 changes: 2 additions & 2 deletions Chapter9/convolutional_networks.tex
Original file line number Diff line number Diff line change
Expand Up @@ -1104,9 +1104,9 @@ \section{\glsentrytext{convolutional_network}的神经科学基础}
\else
\centering
\subfigure{ \label{fig:chap9_feature_detectors_a}
\includegraphics[width=0.4\textwidth]{Chapter9/figures/maxout_kernels.png}}
\subfigure{ \label{fig:chap9_feature_detectors_b}
\includegraphics[width=0.4\textwidth]{Chapter9/figures/s3c_filters.png}}
\subfigure{ \label{fig:chap9_feature_detectors_b}
\includegraphics[width=0.4\textwidth]{Chapter9/figures/maxout_kernels.png}}
\fi
\caption{许多机器学习算法在应用于自然图像时,会学习那些用来检测边缘或边缘的特定颜色的特征。
这些特征检测器使人联想到已知存在于初级视觉皮层中的~\gls{Gabor_function}。
Expand Down
138 changes: 134 additions & 4 deletions acknowledgments.tex
Original file line number Diff line number Diff line change
@@ -1,6 +1,136 @@
% !Mode:: "TeX:UTF-8"
%TODO
\chapter*{致谢}

\chapter*{中文版致谢}
\addcontentsline{toc}{chapter}{致谢}
TODO

首先,我们感谢原作者在本书翻译时给予我们的大力支持。

本书涉及的内容广泛而思想深刻,如果没有众多同学和网友的帮助,我们不可能顺利完成翻译。

我们才疏学浅而受此重任,深知自身水平难以将本书翻译得很准确。
因此我们完成草稿后,将书稿公开于Github,及早接受网友的批评和建议。
以下网友为本书的翻译草稿提供了很多及时的反馈和宝贵的修改意见:
@tttwwy @tankeco @fairmiracle @GageGao @huangpingchun @MaHongP @acgtyrant @yanhuibin315 @Buttonwood @titicacafz @weijy026a @RuiZhang1993 @zymiboxpay @xingkongliang
@oisc @tielei @yuduowu @Qingmu @HC-2016 @xiaomingabc @bengordai @Bojian @JoyFYan @minoriwww @khty2000 @gump88 @zdx3578 @PassStory @imwebson @wlbksy @roachsinai
@Elvinczp @endymecy @9578577 @linzhp @cnscottzheng @germany-zhu @zhangyafeikimi @showgood163 @kangqf @NeutronT @badpoem @kkpoker @Seaball @wheaio @angrymidiao
@ZhiweiYang @corenel @zhaoyu611 @SiriusXDJ @dfcv24 @EmisXXY @FlyingFire @vsooda @friskit-china @poerin @ninesunqian @JiaqiYao @Sofring @wenlei @wizyoung
@imageslr @indam @XuLYC @zhouqingping @freedomRen @runPenguin @piantou

在此期间,我们四位译者再次进行了校对并相互之间也校对了一遍。
然而仅仅通过我们的校对,实在难以发现翻译中存在的问题。
因此,我们邀请一些同学和网友帮助我们校对。
经过他们的校对,本书的翻译质量提升了不少。
我们一一列出,以表示我们由衷的感谢!

\begin{itemize}
\item 第一章(前言): 刘畅、许丁杰、潘雨粟和NeutronT对本章进行了阅读,并对很多语句提出了不少修改建议。林中鹏进行了校对,他提出了很多独到的修改建议。
\item 第二章(线性代数):许丁杰和骆徐圣阅读本章,并修改语句。李若愚进行了校对,提出了很多细心的建议。
\item 第三章(概率与信息论):许丁杰阅读本章,并修改语句。李培炎和何翊卓进行了校对,并修改了很多中文用词,使翻译更加准确。
\item 第四章(数值计算):张亚霏阅读本章,并对其他章节也有提出了一些修改建议。张源源进行了校对,并指出了原文可能存在的问题,非常仔细。
\item 第五章(机器学习基础):郭浩和黄平春阅读本章,并修改语句。李东和林中鹏进行了校对。本章篇幅较长,能够有现在的翻译质量离不开这四位的贡献。
\item 第六章(深度前馈网络):周卫林、林中鹏和张远航阅读本章,并提出修改意见。
\item 第七章(深度学习中的正则化):周柏村进行了非常细心的校对,指出了大量问题,令翻译更加准确。
\item 第八章(深度模型中的优化):房晓宇和吴翔阅读本章。黄平春进行了校对,他提出的很多建议让行文更加流畅易懂。
\item 第九章(卷积网络):赵雨和潘雨粟阅读本章,并润色语句。丁志铭进行了非常仔细的校对,并指出很多翻译问题。
\item 第十章(序列建模:循环和递归网络):刘畅阅读本章。赵雨提供了详细的校对建议,尹瑞清根据他的翻译版本,给我们的版本提出了很多建议。虽然仍存在一些分歧,但我们两个版本的整合,让翻译质量提升很多。
\item 第十二章(应用):潘雨粟进行了校对,在他的校对之前,本章阅读起来比较困难。他提供的修改建议,不仅提高了行文流畅度,还提升了译文的准确度。
\item 第十三章(线性因子模型):贺天行阅读本章,修改语句。杨志伟校对本章,润色大量语句。
\item 第十四章(自编码器):李雨慧和黄平春进行了校对。李雨慧提升了语言的流畅度,黄平春纠正了不少错误,提高了准确性。
\item 第十五章(表示学习):cnscottzheng阅读本章,并修改语句。
\item 第十七章(蒙特卡罗方法):张远航提供了非常细致的校对,后续还校对了一遍,使译文质量大大提升。
\item 第十八章(面对配分函数):吴家楠进行了校对,提升了译文准确性和可读性。
\item 第十九章(近似推断):张远航和张源源进行了校对。这章虽篇幅不大,但内容有深度,译文在两位的帮助下提高了准确度。
\end{itemize}

所有校对的修改建议都保存在Github上,再次感谢以上同学和网友的付出。
经过这五个多月的修改,草稿慢慢变成了初稿。
尽管还有很多问题,但大部分内容是可读的,并且是准确的。
当然目前的翻译仍存在一些没有及时发现的问题,因此翻译也将持续更新,不断修改。
我们非常希望读者能到Github提建议,并且非常欢迎,无论多么小的修改建议,都是非常宝贵的。

此外,我们还要感谢魏太云学长,他与出版社沟通交流,给我们提供了很多排版上的指导。
最后,我们感谢张志华教授的支持。没有老师的帮助,我们也难以完成翻译。

\chapter*{原书致谢}
This book would not have been possible without the contributions of many people.

We would like to thank those who commented on our proposal for the book and helped plan its contents and organization:
Guillaume Alain, Kyunghyun Cho, \c{C}a\u{g}lar G\"ul\c{c}ehre, David Krueger, Hugo Larochelle, Razvan Pascanu and Thomas Roh\'ee.

We would like to thank the people who offered feedback on the content of the book itself. Some offered feedback on many chapters:
Mart\'in Abadi, Guillaume Alain, Ion Androutsopoulos, Fred Bertsch, Olexa Bilaniuk, Ufuk Can Biçici, Matko Bo\v{s}njak, John Boersma, Greg Brockman, Alexandre de Brébisson, Pierre Luc Carrier, Sarath Chandar, Pawel Chilinski, Mark Daoust, Oleg Dashevskii, Laurent Dinh, Stephan Dreseitl, Jim Fan, Miao Fan, Meire Fortunato, Fr\'ed\'eric Francis, Nando de Freitas, \c{C}a\u{g}lar G\"ul\c{c}ehre, Jurgen Van Gael, Javier Alonso Garc\'ia, Jonathan Hunt, Gopi Jeyaram, Chingiz Kabytayev, Lukasz Kaiser, Varun Kanade, Asifullah Khan, Akiel Khan, John King, Diederik P. Kingma, Yann LeCun, Rudolf Mathey, Matías Mattamala, Abhinav Maurya, Kevin Murphy, Oleg Mürk, Roman Novak, Augustus Q. Odena, Simon Pavlik, Karl Pichotta, Eddie Pierce, Kari Pulli, Roussel Rahman, Tapani Raiko, Anurag Ranjan, Johannes Roith, Mihaela Rosca, Halis Sak, César Salgado, Grigory Sapunov, Yoshinori Sasaki, Mike Schuster, Julian Serban, Nir Shabat, Ken Shirriff, Andre Simpelo, Scott Stanley, David Sussillo, Ilya Sutskever, Carles Gelada Sáez, Graham Taylor, Valentin Tolmer, Massimiliano Tomassoli, An Tran, Shubhendu Trivedi, Alexey Umnov, Vincent Vanhoucke, Marco Visentini-Scarzanella, Martin Vita, David Warde-Farley, Dustin Webb, Kelvin Xu, Wei Xue, Ke Yang, Li Yao, Zygmunt Zaj\k{a}c and Ozan \c{C}a\u{g}layan.

We would also like to thank those who provided us with useful feedback on individual chapters:

\begin{itemize}
\item Notation: Zhang Yuanhang.
\item
Chapter 1, Introduction:
Yusuf Akgul, Sebastien Bratieres, Samira Ebrahimi, Charlie Gorichanaz, Brendan Loudermilk, Eric Morris, Cosmin Pârvulescu and Alfredo Solano.
\item Chapter 2, Linear Algebra:
Amjad Almahairi, Nikola Bani\'{c}, Kevin Bennett, Philippe Castonguay, Oscar Chang, Eric Fosler-Lussier, Andrey Khalyavin, Sergey Oreshkov, Istv\'an Petr\'as, Dennis Prangle, Thomas Roh\'ee, Gitanjali Gulve Sehgal, Colby Toland, Alessandro Vitale and Bob Welland.
\item Chapter 3, Probability and Information Theory:
John Philip Anderson, Kai Arulkumaran, Vincent Dumoulin, Rui Fa, Stephan Gouws, Artem Oboturov, Antti Rasmus, Alexey Surkov and Volker Tresp.
\item Chapter 4, Numerical Computation:
Tran Lam AnIan Fischer and Hu Yuhuang.
\item Chapter 5, Machine Learning Basics:
Dzmitry Bahdanau, Justin Domingue, Nikhil Garg, Makoto Otsuka, Bob Pepin, Philip Popien, Emmanuel Rayner, Peter Shepard, Kee-Bong Song, Zheng Sun and Andy Wu.
\item Chapter 6, Deep Feedforward Networks:
Uriel Berdugo, Fabrizio Bottarel, Elizabeth Burl, Ishan Durugkar, Jeff Hlywa, Jong Wook Kim, David Krueger and Aditya Kumar Praharaj.
\item Chapter 7, Regularization for Deep Learning:
Morten Kolbæk, Kshitij Lauria, Inkyu Lee, Sunil Mohan, Hai Phong Phan and Joshua Salisbury.
\item Chapter 8, Optimization for Training Deep Models:
Marcel Ackermann, Peter Armitage, Rowel Atienza, Andrew Brock, Tegan Maharaj, James Martens, Kashif Rasul, Klaus Strobl and Nicholas Turner.
\item Chapter 9, Convolutional Networks:
Mart\'in Arjovsky, Eugene Brevdo, Konstantin Divilov, Eric Jensen, Mehdi Mirza, Alex Paino, Marjorie Sayer, Ryan Stout and Wentao Wu.
\item Chapter 10, Sequence Modeling:
Gökçen Eraslan, Steven Hickson, Razvan Pascanu, Lorenzo von Ritter, Rui Rodrigues, Dmitriy Serdyuk, Dongyu Shi and Kaiyu Yang.
\item Chapter 11, Practical Methodology:
Daniel Beckstein.
\item Chapter 12, Applications:
George Dahl, Vladimir Nekrasov and Ribana Roscher.
\item Chapter 13, Linear Factor Models:
Jayanth Koushik.
\item Chapter 15, Representation Learning:
Kunal Ghosh.
\item Chapter 16, Structured Probabilistic Models for Deep Learning:
Minh Lê and Anton Varfolom.
\item Chapter 18, Confronting the Partition Function:
Sam Bowman.
\item Chapter 19, Approximate Inference:
Yujia Bao.
\item Chapter 20, Deep Generative Models:
Nicolas Chapados, Daniel Galvez, Wenming Ma, Fady Medhat, Shakir Mohamed and Gr\'egoire Montavon.
\item Bibliography:
Lukas Michelbacher and Leslie N. Smith.
\end{itemize}
% CHECK: make sure the chapters are still in order

We also want to thank those who allowed us to reproduce images, figures or data from
their publications.
We indicate their contributions
in the figure captions throughout the text.
% David Warde-Farley,
% Matthew D. Zeiler,
% Rob Fergus,
% Nicolas Chapados,
% Razvan Pascanu,
% James Bergstra,
% Dumitru Erhan,
% Emily Denton
% and Soumith Chintala.

We would like to thank Lu Wang for writing pdf2htmlEX, which we used
to make the web version of the book, and for offering support to
improve the quality of the resulting HTML.

We would like to thank Ian's wife Daniela Flori Goodfellow for
patiently supporting Ian during the writing of the book as well as for
help with proofreading.

We would like to thank the Google Brain team for providing an
intellectual environment where Ian could devote a tremendous amount of
time to writing this book and receive feedback and guidance from
colleagues. We would especially like to thank Ian's former manager,
Greg Corrado, and his current manager, Samy Bengio, for their support
of this project. Finally, we would like to thank Geoffrey Hinton for
encouragement when writing was difficult.
4 changes: 2 additions & 2 deletions docs/_posts/2016-12-05-Chapter5_machine_learning_basics.md
Original file line number Diff line number Diff line change
Expand Up @@ -250,7 +250,7 @@ Iris(鸢尾花卉)数据集~{cite?}是统计学家和机器学习研究者
这意味着我们可以将该数据集表示为设计矩阵 $\MX\in\SetR^{150\times 4}$,其中$X_{i,1}$表示第$i$个植物的萼片长度,$X_{i,2}$表示第$i$个植物的萼片宽度等等。
我们在本书中描述的大部分学习算法都是讲述它们是如何运行在设计矩阵数据集上的。

当然,每一个样本都能表示成向量,并且这些向量的大小相同,才能将一个数据集表示成设计矩阵。
当然,每一个样本都能表示成向量,并且这些向量的维度相同,才能将一个数据集表示成设计矩阵。
这一点并非永远可能。
例如,你有不同宽度和高度的照片的集合,那么不同的照片将会包含不同数量的像素。
因此不是所有的照片都可以表示成相同长度的向量。
Expand Down Expand Up @@ -545,7 +545,7 @@ VC\,维定义为该分类器能够分类的训练样本的最大数目。
当需要为测试点$\Vx$分类时,模型会查询训练集中离该点最近的点,并返回相关的回归目标。
换言之,$\hat{y}=y_i$其中$i=\argmin \norm{\MX_{i,:}-\Vx}_2^2$。
该算法也可以扩展成$L^2$范数以外的距离度量,例如学成距离度量{cite?}。
在有多个最近向量存在的情况下,如果允许该算法通过平均所有最近的$\MX_{i,:}$对应的$y_i$来打破平局,那么该算法会在任意回归数据集上达到最小可能的训练误差(如果存在两个相同的输入对应不同的输出,那么训练误差可能会大于零)。
在最近向量不唯一的情况下,如果允许算法对所有离$\Vx$最近的$\MX_{i,:}$关联的$y_i$求平均,那么该算法会在任意回归数据集上达到最小可能的训练误差(如果存在两个相同的输入对应不同的输出,那么训练误差可能会大于零)。

最后,我们也可以将参数学习算法嵌入另一个增加参数数目的算法来创建非参数学习算法。
例如,我们可以想象这样一个算法,外层循环调整多项式的次数,内层循环通过线性回归学习模型。
Expand Down
2 changes: 1 addition & 1 deletion docs/_posts/2016-12-07-Chapter7_regularization.md
Original file line number Diff line number Diff line change
Expand Up @@ -717,7 +717,7 @@ softmax函数~永远无法真正预测0概率或1概率,因此它会继续学
\ifOpenSource
\centerline{\includegraphics{figure.pdf}}
\else
\centerline{\includegraphics[width=0.8\textwidth]{Chapter7/figures/reg_l1_vs_l2_mistake}}
\centerline{\includegraphics[width=0.8\textwidth]{Chapter7/figures/reg_early_stop_vs_l2}}
\fi
\caption{提前终止效果的示意图。
\emph{(左)}实线轮廓线表示负对数似然的轮廓。
Expand Down
2 changes: 1 addition & 1 deletion docs/_posts/2016-12-09-Chapter9_convolutional_networks.md
Original file line number Diff line number Diff line change
Expand Up @@ -1089,8 +1089,8 @@ Gabor函数描述在图像中的2维点处的权重。我们可以认为图像
\centerline{\includegraphics{figure.pdf}}
\else
\centering
\includegraphics[width=0.4\textwidth]{Chapter9/figures/maxout_kernels.png}}
\includegraphics[width=0.4\textwidth]{Chapter9/figures/s3c_filters.png}}
\includegraphics[width=0.4\textwidth]{Chapter9/figures/maxout_kernels.png}}
\fi
\caption{许多机器学习算法在应用于自然图像时,会学习那些用来检测边缘或边缘的特定颜色的特征。
这些特征检测器使人联想到已知存在于初级视觉皮层中的~Gabor函数。
Expand Down
4 changes: 2 additions & 2 deletions docs/_posts/2016-12-11-Chapter11_practical_methodology.md
Original file line number Diff line number Diff line change
Expand Up @@ -427,8 +427,8 @@ Dropout\,比率 & 降低 & 较少地丢弃单元可以更多地让单元彼此"
\centerline{\includegraphics{figure.pdf}}
\else
\begin{tabular}{cc}
\includegraphics[width=0.4\textwidth]{Chapter11/figures/grid} &
\includegraphics[width=0.4\textwidth]{Chapter11/figures/random}
\includegraphics[width=0.35\textwidth]{Chapter11/figures/grid} &
\includegraphics[width=0.35\textwidth]{Chapter11/figures/random}
\end{tabular}
\fi
\caption{网格搜索和随机搜索的比较。
Expand Down
4 changes: 2 additions & 2 deletions docs/_posts/2016-12-12-Chapter12_applications.md
Original file line number Diff line number Diff line change
Expand Up @@ -456,10 +456,10 @@ sphering~通常被称为白化。
\else
\centering
\begin{tabular}{ccc}
\includegraphics[width=.3\figwidth]{Chapter12/figures/src0.jpg} &
\includegraphics[width=.3\figwidth]{Chapter12/figures/gray0.jpg} &
\includegraphics[width=.3\figwidth]{Chapter12/figures/gcn0.jpg} &
\includegraphics[width=.3\figwidth]{Chapter12/figures/lcn0.jpg} \\
\includegraphics[width=.3\figwidth]{Chapter12/figures/src1.jpg} & % ?? may be problem
\includegraphics[width=.3\figwidth]{Chapter12/figures/gray1.jpg} & % ?? may be problem
\includegraphics[width=.3\figwidth]{Chapter12/figures/gcn1.jpg} &
\includegraphics[width=.3\figwidth]{Chapter12/figures/lcn1.jpg}\\
Input image & GCN & LCN
Expand Down
4 changes: 2 additions & 2 deletions docs/_posts/2016-12-15-Chapter15_representation_learning.md
Original file line number Diff line number Diff line change
Expand Up @@ -578,8 +578,8 @@ $\Vh_{\Vx}$空间中的相似性度量表示$\Vx$空间中任意点对之间的
\else
\begin{tabular}{cc}
输入 & 重构 \\
\includegraphics[width=0.45\textwidth]{Chapter15/figures/ping_pong_input} &
\includegraphics[width=0.45\textwidth]{Chapter15/figures/ping_pong_reconstruction}
\includegraphics[width=0.4\textwidth]{Chapter15/figures/ping_pong_input} &
\includegraphics[width=0.4\textwidth]{Chapter15/figures/ping_pong_reconstruction}
\end{tabular}
\fi
\caption{机器人任务上,基于均方误差训练的自编码器不能重构乒乓球。
Expand Down
1 change: 0 additions & 1 deletion docs/_posts/2016-12-17-Chapter17_monte_carlo_methods.md
Original file line number Diff line number Diff line change
Expand Up @@ -459,7 +459,6 @@ Gibbs采样混合得很慢,因为每次更新仅仅一个变量很难跨越不
但是对于吉布斯链来说从分布的一个峰值转移到另一个仍然是很困难的,比如说改变数字。
\emph{(右)}从生成式对抗网络中抽出的连续原始样本。
因为原始采样生成的样本之间互相独立,所以不存在混合问题。
{译者注:原书此处左右搞反了。}}
\end{figure}
<!-- % 593 end -->

Expand Down

0 comments on commit 4f9fca0

Please sign in to comment.