https://blog.youkuaiyun.com/cppjava_zh-cnhttps://blog.youkuaiyun.com/https://blog.youkuaiyun.com/cppjava_/article/details/77188391https://blog.youkuaiyun.com/cppjava_/article/details/77188391cppjava_Tue, 15 Aug 2017 14:11:50 +0800https://blog.youkuaiyun.com/cppjava_/article/details/76820427https://blog.youkuaiyun.com/cppjava_/article/details/76820427cppjava_Mon, 07 Aug 2017 11:57:48 +0800https://blog.youkuaiyun.com/cppjava_/article/details/72553243https://blog.youkuaiyun.com/cppjava_/article/details/72553243cppjava_Fri, 19 May 2017 19:20:20 +0800https://blog.youkuaiyun.com/cppjava_/article/details/72331524https://blog.youkuaiyun.com/cppjava_/article/details/72331524cppjava_Tue, 16 May 2017 20:25:22 +0800https://blog.youkuaiyun.com/cppjava_/article/details/71249254https://blog.youkuaiyun.com/cppjava_/article/details/71249254cppjava_Sat, 06 May 2017 10:54:48 +0800卷积层->激励层->卷积层->激励层。。。 数据输入层（数据预处理）：三种方法：去均值（即0均值化，CNN常用，训练集所有像素值减去均值，把输入数据各个维度中心化到0，测试集也减相同的均值）；归一化（幅度归一化到同样的范围）；PCA/白化（降维，白化是对数据每个特征轴上的幅度归一化）。 去均值的目的：（1）数据有过大的均值可能导致参数的梯度过大，在梯度回]]>https://blog.youkuaiyun.com/cppjava_/article/details/71249209https://blog.youkuaiyun.com/cppjava_/article/details/71249209cppjava_Sat, 06 May 2017 10:48:02 +0800https://blog.youkuaiyun.com/cppjava_/article/details/68941436https://blog.youkuaiyun.com/cppjava_/article/details/68941436cppjava_Sat, 01 Apr 2017 10:07:05 +0800https://blog.youkuaiyun.com/cppjava_/article/details/68922113https://blog.youkuaiyun.com/cppjava_/article/details/68922113cppjava_Fri, 31 Mar 2017 10:05:50 +0800https://blog.youkuaiyun.com/cppjava_/article/details/68921869https://blog.youkuaiyun.com/cppjava_/article/details/68921869cppjava_Fri, 31 Mar 2017 09:45:33 +0800https://blog.youkuaiyun.com/cppjava_/article/details/68485565https://blog.youkuaiyun.com/cppjava_/article/details/68485565cppjava_Thu, 30 Mar 2017 12:43:01 +0800https://blog.youkuaiyun.com/cppjava_/article/details/68485536https://blog.youkuaiyun.com/cppjava_/article/details/68485536cppjava_Thu, 30 Mar 2017 12:40:47 +0800https://blog.youkuaiyun.com/cppjava_/article/details/68066680https://blog.youkuaiyun.com/cppjava_/article/details/68066680cppjava_Wed, 29 Mar 2017 16:41:10 +0800https://blog.youkuaiyun.com/cppjava_/article/details/68062990https://blog.youkuaiyun.com/cppjava_/article/details/68062990cppjava_Wed, 29 Mar 2017 12:10:45 +0800>n; int *a=new int[n]; for(int i=0;i { cin>>a[i]; } ....    d]]>https://blog.youkuaiyun.com/cppjava_/article/details/68060439https://blog.youkuaiyun.com/cppjava_/article/details/68060439cppjava_Wed, 29 Mar 2017 10:19:21 +0800=+1,负类样本y =wTx+ b1右边，负类样本位于H2左边，所以原始的代价函数就为： 函数间隔（functional margin）: 几何间隔（geometric margi]]>https://blog.youkuaiyun.com/cppjava_/article/details/67668230https://blog.youkuaiyun.com/cppjava_/article/details/67668230cppjava_Tue, 28 Mar 2017 22:30:05 +0800 (1)  批处理梯度下降法：每次对全部训练数据进行求导后更新参数，能求得最优解； (2)  增量梯度下降法]]>https://blog.youkuaiyun.com/cppjava_/article/details/67663085https://blog.youkuaiyun.com/cppjava_/article/details/67663085cppjava_Tue, 28 Mar 2017 22:20:52 +0800https://blog.youkuaiyun.com/cppjava_/article/details/53898559https://blog.youkuaiyun.com/cppjava_/article/details/53898559cppjava_Tue, 27 Dec 2016 19:56:15 +0800https://blog.youkuaiyun.com/cppjava_/article/details/53539868https://blog.youkuaiyun.com/cppjava_/article/details/53539868cppjava_Fri, 09 Dec 2016 15:49:20 +0800https://blog.youkuaiyun.com/cppjava_/article/details/53432080https://blog.youkuaiyun.com/cppjava_/article/details/53432080cppjava_Fri, 02 Dec 2016 10:21:24 +0800