add: AdaBN

Author: jindongwang

README.md

@@ -56,6 +56,7 @@
 - 在[这个issue](https://github.com/jindongwang/transferlearning-tutorial/issues/1)下留言你的Github账号和邮箱，我将你添加到协作者中
 - 直接fork，然后将你的修改提交pull request
 - 如果不熟悉git，可直接下载本目录，然后将你修改的部分发给我(jindongwang@outlook.com)
+- 有任何问题，均可以提交issue
 
 然后在下面的贡献者信息中加入自己的信息。
 

src/chaps/deep.tex

@@ -292,6 +292,19 @@ \subsubsection{核心方法}
 	\label{fig-deep-jan}
 \end{figure}
 
+\textbf{5. AdaBN}
+
+与上述工作选择在已有网络层中增加适配层不同的是，北京大学的Haoyang Li和图森科技的Naiyan Wang等人提出了AdaBN(Adaptive Batch Normalization)~\cite{li2018adaptive}，通过在归一化层加入统计特征的适配，从而完成迁移。
+
+\begin{figure}[htbp]
+	\centering
+	\includegraphics[scale=0.38]{./figures/fig-deep-adabn.pdf}
+	\caption{AdaBN方法示意图}
+	\label{fig-deep-adabn}
+\end{figure}
+
+AdaBN对比其他方法，实现相当简单。并且，方法本身不带有任何额外的参数。在许多公开数据集上都取得了很好的效果。
+
 \subsubsection{小结}
 
 基于深度网络进行迁移学习，其核心在于，找到网络需要进行自适应的层，并且对这些导加上自适应的损失度量。越来越多的研究者开始使用深度网络进行迁移学习~\cite{long2016deep,zhuo2017deep,zhuang2015supervised,sun2016deep,wei2016deep,luo2017label}。在这其中，几乎绝大多数方法都采用了卷积神经网络，在已训练好的模型(如AlexNet、Inception、GoogLeNet、Resnet等)上进行迁移。

src/figures/fig-deep-adabn.pdf

@@ -62,22 +62,22 @@
 \contentsline {subsection}{\numberline {9.3}深度网络自适应}{42}{subsection.9.3}
 \contentsline {subsubsection}{\numberline {9.3.1}基本思路}{42}{subsubsection.9.3.1}
 \contentsline {subsubsection}{\numberline {9.3.2}核心方法}{43}{subsubsection.9.3.2}
-\contentsline {subsubsection}{\numberline {9.3.3}小结}{47}{subsubsection.9.3.3}
-\contentsline {subsection}{\numberline {9.4}深度对抗网络迁移}{47}{subsection.9.4}
+\contentsline {subsubsection}{\numberline {9.3.3}小结}{48}{subsubsection.9.3.3}
+\contentsline {subsection}{\numberline {9.4}深度对抗网络迁移}{48}{subsection.9.4}
 \contentsline {subsubsection}{\numberline {9.4.1}基本思路}{48}{subsubsection.9.4.1}
 \contentsline {subsubsection}{\numberline {9.4.2}核心方法}{48}{subsubsection.9.4.2}
-\contentsline {subsubsection}{\numberline {9.4.3}小结}{50}{subsubsection.9.4.3}
-\contentsline {section}{\numberline {10}上手实践}{51}{section.10}
-\contentsline {section}{\numberline {11}迁移学习前沿}{57}{section.11}
-\contentsline {subsection}{\numberline {11.1}机器智能与人类经验结合迁移}{57}{subsection.11.1}
-\contentsline {subsection}{\numberline {11.2}传递式迁移学习}{57}{subsection.11.2}
-\contentsline {subsection}{\numberline {11.3}终身迁移学习}{58}{subsection.11.3}
-\contentsline {subsection}{\numberline {11.4}在线迁移学习}{59}{subsection.11.4}
-\contentsline {subsection}{\numberline {11.5}迁移强化学习}{60}{subsection.11.5}
-\contentsline {subsection}{\numberline {11.6}迁移学习的可解释性}{60}{subsection.11.6}
-\contentsline {section}{\numberline {12}总结语}{61}{section.12}
-\contentsline {section}{\numberline {13}附录}{62}{section.13}
-\contentsline {subsection}{\numberline {13.1}迁移学习相关的期刊和会议}{62}{subsection.13.1}
-\contentsline {subsection}{\numberline {13.2}迁移学习研究学者}{62}{subsection.13.2}
-\contentsline {subsection}{\numberline {13.3}迁移学习资源汇总}{65}{subsection.13.3}
-\contentsline {subsection}{\numberline {13.4}迁移学习常用算法及数据资源}{66}{subsection.13.4}
+\contentsline {subsubsection}{\numberline {9.4.3}小结}{51}{subsubsection.9.4.3}
+\contentsline {section}{\numberline {10}上手实践}{52}{section.10}
+\contentsline {section}{\numberline {11}迁移学习前沿}{58}{section.11}
+\contentsline {subsection}{\numberline {11.1}机器智能与人类经验结合迁移}{58}{subsection.11.1}
+\contentsline {subsection}{\numberline {11.2}传递式迁移学习}{58}{subsection.11.2}
+\contentsline {subsection}{\numberline {11.3}终身迁移学习}{59}{subsection.11.3}
+\contentsline {subsection}{\numberline {11.4}在线迁移学习}{60}{subsection.11.4}
+\contentsline {subsection}{\numberline {11.5}迁移强化学习}{61}{subsection.11.5}
+\contentsline {subsection}{\numberline {11.6}迁移学习的可解释性}{61}{subsection.11.6}
+\contentsline {section}{\numberline {12}总结语}{62}{section.12}
+\contentsline {section}{\numberline {13}附录}{63}{section.13}
+\contentsline {subsection}{\numberline {13.1}迁移学习相关的期刊和会议}{63}{subsection.13.1}
+\contentsline {subsection}{\numberline {13.2}迁移学习研究学者}{63}{subsection.13.2}
+\contentsline {subsection}{\numberline {13.3}迁移学习资源汇总}{66}{subsection.13.3}
+\contentsline {subsection}{\numberline {13.4}迁移学习常用算法及数据资源}{67}{subsection.13.4}

src/main.aux

@@ -2191,4 +2191,14 @@ @misc{xinhua
 	howpublished="\url{http://mp.weixin.qq.com/s?__biz=MjM5ODYzNzAyMQ==\&mid=2651933920\&idx=1\\&sn=ae2866bd12000f1644eae1094497837e}",
 	year={2016}
 
-}
+}
+
+@article{li2018adaptive,
+	title={Adaptive Batch Normalization for practical domain adaptation},
+	author={Li, Yanghao and Wang, Naiyan and Shi, Jianping and Hou, Xiaodi and Liu, Jiaying},
+	journal={Pattern Recognition},
+	volume={80},
+	pages={109--117},
+	year={2018},
+	publisher={Elsevier}
+}