【读书1】【2017】MATLAB与深度学习——多层神经网络的训练(1)

1986年，反向传播算法的引入最终解决了多层神经网络的训练问题。

In 1986, the introduction of theback-propagation algorithm finally solved the training problem of themulti-layer neural network.

反向传播算法的意义在于，它提供了一种确定隐藏节点误差的系统方法。

The significance of the back-propagationalgorithm was that it provided a systematic method to determine the error ofthe hidden nodes.

一旦能够确定隐藏层输出误差，则可以应用增量规则来调整权重。

Once the hidden layer errors aredetermined, the delta rule is applied to adjust the weights.

如图3-1所示。

See Figure 3-1.

图3-1 反向传播算法示意图Illustration ofback-propagation

神经网络的输入数据通过输入层、隐藏层和输出层进行传播。

The input data of the neural networktravels through the input layer, hidden layer, and output layer.

相比之下，在反向传播算法中，输出误差从输出层开始，向相反方向移动，直到到达输入层的下一级隐藏层。

In contrast, in the back-propagationalgorithm, the output error starts from the output layer and moves backwarduntil it reaches the right next hidden layer to the input layer.

这个过程被称为反向传播，因为它类似于向后传播的输出误差。

This process is called back propagation, asit resembles an output error propagating backward.

即使在反向传播中，信号仍然流过所有的连接线，并且被乘以相应的权值。

Even in back-propagation, the signal stillflows through the connecting lines and the weights are multiplied.

唯一的区别是输入和输出信号的流动方向相反。

The only difference is that the input andoutput signals flow in opposite directions.

反向传播算法（Back-Propagation Algorithm）

本节以简单的多层神经网络为例说明反向传播算法。

This section explains the back-propagationalgorithm using an example of the simple multi-layer neural network.

考虑一个神经网络，它由输入和输出的两个节点和一个隐藏层组成，隐藏层也具有两个节点。

Consider a neural network that consists oftwo nodes for both the input and output and a hidden layer, which has two nodesas well.

为了研究方便，我们将省略了偏置参数，即偏置设为0。

We will omit the bias for convenience.

示例神经网络如图3-2所示，其中上标为层指示符。

The example neural network is shown inFigure 3-2, where the superscript describes the layer indicator.

图3-2 神经网络的输入层、输出层以及隐藏层都具有两个节点Neuralnetwork that consists of two nodes for the input and output and a hidden layer,which has two nodes

为了获得输出误差，我们首先需要计算出输入数据对应的神经网络输出。

In order to obtain the output error, wefirst need the neural network’s output from the input data.

让我们试试看。

Let’s try.

由于示例网络具有单个隐藏层，所以在进行输出计算之前需要对两个输入数据操作。

As the example network has a single hiddenlayer, we need two input data manipulations before the output calculation isprocessed.

首先，计算隐藏节点的加权和为：

First, the weighted sum of the hidden nodeis calculated as:

更多精彩文章请关注微信号：