war3gu的博客

softmax e的N次方，然后除以和 square error 方差 cross entroy 交叉熵，解决方差梯度太小，回归太慢的问题 Adagrad 计算learning rate RMSProp 计算learning rate Momentum 修正梯度方向 Adam RMSProp+Momentum dropout ...

if we just keep stacking layers, is the network get better every time? 如果我们不停堆积层次，网络是否每次都会变得更好 Create an interface, which takes as input the business object, and as output a type of string 这是一个倒装句， ta...