交叉熵函数与方程3.9的二次函数之间的主要区别是它的增长呈现为几何增长。
The primary difference of the cross entropyfunction from the quadratic function of Equation 3.9 is its geometric increase.
换句话说,交叉熵函数对误差更敏感。
In other words, the cross entropy functionis much more sensitive to the error.
出于这个原因,通常从交叉熵函数推导出的学习规则通常能够产生更好的性能。
For this reason, the learning rules derivedfrom the cross entropy function are generally known to yield betterperformance.
建议使用交叉熵驱动的学习规则,除了回归等不可避免的情况。
It is recommended that you use the crossentropy driven learning rules except for inevitable cases such as theregression.
由于代价函数的选择影响学习规则,即反向传播算法的公式,因此我们对代价函数进行了长篇幅的介绍。
We had a long introduction to the costfunction because the selection of the cost function affects the learning rule,i.e., the formula of the back-propagation algorithm.
具体而言,输出节点上的增量计算略有变化。
Specifically, the calculation of the deltaat the output node changes slightly.
以下步骤详细说明了使用交叉熵驱动的反向传播算法在输出节点处采用sigmoid激活函数训练神经网络的过程。
The following steps detail the procedure intraining the neural network with the sigmoid activation function at the outputnode using the cross entropy driven back-propagation algorithm.
-
用适当的值初始化神经网络的权值。
-
Initialize the neural network’s weightswith adequate values.
-
将训练数据即{输入,正确输出}输入神经网络并获得相应的输出。
-
Enter the input of the training data {input, correct output } to the neural network and obtain the output.
其它步骤和内容都是完全相同的。
Everything else remains the same.
外部的差异似乎都微不足道。
On the outside, the difference seemsinsignificant.
然而,它却包含了基于优化理论的代价函数的庞大课题。
However, it contains the huge topic of thecost function based on the optimization theory.
深度学习的神经网络训练方法大多采用交叉熵驱动的学习规则。
Most of the neural network trainingapproaches of Deep Learning employ the cross entropy-driven learning rules.
这是由于这些方法的学习速度快且性能优越。
This is due to their superior learning rateand performance.
——本文译自Phil Kim所著的《Matlab Deep Learning》
更多精彩文章请关注微信号: