<?xml version="1.0" encoding="utf-8" ?><rss version="2.0"><channel><title><![CDATA[NeuFace的博客]]></title><description><![CDATA[]]></description><link>https://blog.youkuaiyun.com/NeuFace</link><language>zh-cn</language><generator>https://blog.youkuaiyun.com/</generator><copyright><![CDATA[Copyright &copy; NeuFace]]></copyright><item><title><![CDATA[Training Very Deep Networks]]></title><link>https://blog.youkuaiyun.com/NeuFace/article/details/50485066</link><guid>https://blog.youkuaiyun.com/NeuFace/article/details/50485066</guid><author>NeuFace</author><pubDate>Fri, 08 Jan 2016 20:08:04 +0800</pubDate><description><![CDATA[这篇文章提出一种训练深层网络的训练结构-highway，主要的来源于LSTM中的阀门开关的思想。highway的提出使得可以使用梯度下降可以直接训练较深的卷积神经网络。
plain网络前向过程为：



其中H为非线性激活函数。对于highway网络来说在上述的基础上又引入了两个非线性变换T和C，则highway为：


其中T为transform gate，C为carry gat]]></description><category></category></item><item><title><![CDATA[Deep Residual Learning for Image Recognition]]></title><link>https://blog.youkuaiyun.com/NeuFace/article/details/50484823</link><guid>https://blog.youkuaiyun.com/NeuFace/article/details/50484823</guid><author>NeuFace</author><pubDate>Fri, 08 Jan 2016 18:57:25 +0800</pubDate><description><![CDATA[Deep Residual Learning for Image Recognition中提出的152层网络结构在ILSRC 2015获得第一名，集成后的网络在ImageNet分类错误率3.57%，作者微软研究院的何凯明等。


        上图中两条曲线分别为20层和56层的“plain”网络在训练集和测试集上的错误率，这里的“plain”是指一般的无环的“直流”卷积网络。按照以往的经]]></description><category></category></item><item><title><![CDATA[Boosting & AdaBoost算法]]></title><link>https://blog.youkuaiyun.com/NeuFace/article/details/50423783</link><guid>https://blog.youkuaiyun.com/NeuFace/article/details/50423783</guid><author>NeuFace</author><pubDate>Tue, 29 Dec 2015 09:06:59 +0800</pubDate><description><![CDATA[提升和自适应提升方法
Boosting & AdaBoost
1 算法起源
   在机器学习领域中，关键问题就是如何利用观测数据通过学习得到精确估计。但是，在实际应用领域中，构造一个高精度的估计几乎是不可能的。Boosting就给出了一个解决上述问很好的思路。Boosting是一种提高任意给定学习算法准确度的方法。它的思想起源于 valiant在1984年提出的
PAC ( Probabl]]></description><category></category></item><item><title><![CDATA[训练深度网络的梯度弥散及梯度膨胀问题]]></title><link>https://blog.youkuaiyun.com/NeuFace/article/details/50411976</link><guid>https://blog.youkuaiyun.com/NeuFace/article/details/50411976</guid><author>NeuFace</author><pubDate>Sun, 27 Dec 2015 12:32:40 +0800</pubDate><description><![CDATA[在深度网络中，不同层的学习速度差异很大。尤其是在网络后面层学习的情况很好的时候，先前的层常常会在训练时停止不变，基本上学不到东西，这些原因是与基于梯度的学习方法有关。
我们用MNIST数字分类问题作为研究和实验的对象：
这个网络有784个输入神经元，对应于图片28*28=784个像素点，我们设置隐藏神经元为30个，输出层为10个神经元，对应于MNIST的10个数字（0~9），我们训练30epo]]></description><category></category></item><item><title><![CDATA[训练深度模型的优化问题（十九）]]></title><link>https://blog.youkuaiyun.com/NeuFace/article/details/50405437</link><guid>https://blog.youkuaiyun.com/NeuFace/article/details/50405437</guid><author>NeuFace</author><pubDate>Fri, 25 Dec 2015 21:52:16 +0800</pubDate><description><![CDATA[训练深度模型的优化问题，参考Bengio《Deep Learning》]]></description><category></category></item><item><title><![CDATA[训练深度模型的优化问题（十八）]]></title><link>https://blog.youkuaiyun.com/NeuFace/article/details/50405434</link><guid>https://blog.youkuaiyun.com/NeuFace/article/details/50405434</guid><author>NeuFace</author><pubDate>Fri, 25 Dec 2015 21:50:57 +0800</pubDate><description><![CDATA[训练深度模型的优化问题，参考Bengio《Deep Learning》]]></description><category></category></item><item><title><![CDATA[训练深度模型的优化问题（十七）]]></title><link>https://blog.youkuaiyun.com/NeuFace/article/details/50405423</link><guid>https://blog.youkuaiyun.com/NeuFace/article/details/50405423</guid><author>NeuFace</author><pubDate>Fri, 25 Dec 2015 21:50:00 +0800</pubDate><description><![CDATA[训练深度模型的优化问题，参考Bengio《Deep Learning》]]></description><category></category></item><item><title><![CDATA[训练深度模型的优化问题（十六）]]></title><link>https://blog.youkuaiyun.com/NeuFace/article/details/50405407</link><guid>https://blog.youkuaiyun.com/NeuFace/article/details/50405407</guid><author>NeuFace</author><pubDate>Fri, 25 Dec 2015 21:47:35 +0800</pubDate><description><![CDATA[训练深度模型的优化问题，参考Bengio《Deep Learning》]]></description><category></category></item><item><title><![CDATA[训练深度模型的优化问题（十五）]]></title><link>https://blog.youkuaiyun.com/NeuFace/article/details/50405398</link><guid>https://blog.youkuaiyun.com/NeuFace/article/details/50405398</guid><author>NeuFace</author><pubDate>Fri, 25 Dec 2015 21:43:03 +0800</pubDate><description><![CDATA[训练深度模型的优化问题，参考Bengio《Deep Learning》]]></description><category></category></item><item><title><![CDATA[训练深度模型的优化问题（十四）]]></title><link>https://blog.youkuaiyun.com/NeuFace/article/details/50405390</link><guid>https://blog.youkuaiyun.com/NeuFace/article/details/50405390</guid><author>NeuFace</author><pubDate>Fri, 25 Dec 2015 21:41:49 +0800</pubDate><description><![CDATA[训练深度模型的优化问题，参考Bengio《Deep Learning》]]></description><category></category></item><item><title><![CDATA[训练深度模型的优化问题（十三）]]></title><link>https://blog.youkuaiyun.com/NeuFace/article/details/50405387</link><guid>https://blog.youkuaiyun.com/NeuFace/article/details/50405387</guid><author>NeuFace</author><pubDate>Fri, 25 Dec 2015 21:40:26 +0800</pubDate><description><![CDATA[训练深度模型的优化问题，参考Bengio《Deep Learning》]]></description><category></category></item><item><title><![CDATA[训练深度模型的优化问题（十二）]]></title><link>https://blog.youkuaiyun.com/NeuFace/article/details/50405365</link><guid>https://blog.youkuaiyun.com/NeuFace/article/details/50405365</guid><author>NeuFace</author><pubDate>Fri, 25 Dec 2015 21:38:28 +0800</pubDate><description><![CDATA[训练深度模型的优化问题，参考Bengio《Deep Learning》]]></description><category></category></item><item><title><![CDATA[训练深度模型的优化问题（十一）]]></title><link>https://blog.youkuaiyun.com/NeuFace/article/details/50405308</link><guid>https://blog.youkuaiyun.com/NeuFace/article/details/50405308</guid><author>NeuFace</author><pubDate>Fri, 25 Dec 2015 21:32:13 +0800</pubDate><description><![CDATA[训练深度模型的优化问题，参考Bengio《Deep Learning》]]></description><category></category></item><item><title><![CDATA[训练深度模型的优化问题（十）]]></title><link>https://blog.youkuaiyun.com/NeuFace/article/details/50405260</link><guid>https://blog.youkuaiyun.com/NeuFace/article/details/50405260</guid><author>NeuFace</author><pubDate>Fri, 25 Dec 2015 21:13:32 +0800</pubDate><description><![CDATA[训练深度模型的优化问题，参考Bengio《Deep Learning》]]></description><category></category></item><item><title><![CDATA[训练深度模型的优化问题（九）]]></title><link>https://blog.youkuaiyun.com/NeuFace/article/details/50405257</link><guid>https://blog.youkuaiyun.com/NeuFace/article/details/50405257</guid><author>NeuFace</author><pubDate>Fri, 25 Dec 2015 21:01:41 +0800</pubDate><description><![CDATA[训练深度模型的优化问题，参考Bengio《Deep Learning》]]></description><category></category></item><item><title><![CDATA[训练深度模型的优化问题（八）]]></title><link>https://blog.youkuaiyun.com/NeuFace/article/details/50405241</link><guid>https://blog.youkuaiyun.com/NeuFace/article/details/50405241</guid><author>NeuFace</author><pubDate>Fri, 25 Dec 2015 20:59:28 +0800</pubDate><description><![CDATA[训练深度模型的优化问题，参考Bengio《Deep Learning》]]></description><category></category></item><item><title><![CDATA[训练深度模型的优化问题（七）]]></title><link>https://blog.youkuaiyun.com/NeuFace/article/details/50405236</link><guid>https://blog.youkuaiyun.com/NeuFace/article/details/50405236</guid><author>NeuFace</author><pubDate>Fri, 25 Dec 2015 20:56:27 +0800</pubDate><description><![CDATA[训练深度模型的优化问题，参考Bengio《Deep Learning》]]></description><category></category></item><item><title><![CDATA[训练深度模型的优化问题（六）]]></title><link>https://blog.youkuaiyun.com/NeuFace/article/details/50405226</link><guid>https://blog.youkuaiyun.com/NeuFace/article/details/50405226</guid><author>NeuFace</author><pubDate>Fri, 25 Dec 2015 20:54:34 +0800</pubDate><description><![CDATA[训练深度模型的优化问题，参考Bengio《Deep Learning》]]></description><category></category></item><item><title><![CDATA[训练深度模型的优化问题（五）]]></title><link>https://blog.youkuaiyun.com/NeuFace/article/details/50405214</link><guid>https://blog.youkuaiyun.com/NeuFace/article/details/50405214</guid><author>NeuFace</author><pubDate>Fri, 25 Dec 2015 20:53:15 +0800</pubDate><description><![CDATA[训练深度模型的优化问题，参考Bengio《Deep Learning》]]></description><category></category></item><item><title><![CDATA[训练深度模型的优化问题（四）]]></title><link>https://blog.youkuaiyun.com/NeuFace/article/details/50405201</link><guid>https://blog.youkuaiyun.com/NeuFace/article/details/50405201</guid><author>NeuFace</author><pubDate>Fri, 25 Dec 2015 20:49:03 +0800</pubDate><description><![CDATA[训练深度模型的优化问题]]></description><category></category></item></channel></rss>