【读书1】【2017】MATLAB与深度学习——神经网络分层(1)

图2-3 节点分层结构示意图A layered structure ofnodes

图2-3中的方形节点组称为输入层。

The group of square nodes in Figure 2-3 iscalled the input layer.

输入层的节点仅仅充当将输入信号发送到下一节点的通道。

The nodes of the input layer merely act asthe passage that transmits the input signals to the next nodes.

因此，它们不计算加权和与激活函数。

Therefore, they do not calculate theweighted sum and activation function.

这是它们由正方形表示，并区别于其它圆形节点的原因。

This is the reason that they are indicatedby squares and distinguished from the other circular nodes.

相反，最右边的节点组被称为输出层。

In contrast, the group of the rightmostnodes is called the output layer.

从这些节点输出神经网络的最终结果。

The output from these nodes becomes thefinal result of the neural network.

输入层和输出层之间的分层称为隐藏层。

The layers in between the input and outputlayers are called hidden layers.

它们之所以被如此命名，是因为它们不能从神经网络外部访问。

They are given this name because they arenot accessible from the outside of the neural network.

神经网络已经从一个简单的体系结构发展到一个越来越复杂的结构。

The neural network has been developed froma simple architecture to a more and more complex structure.

最初，神经网络的先驱只有一个非常简单的结构，只有输入和输出层，称为单层神经网络。

Initially, neural network pioneers had avery simple architecture with only input and output layers, which are calledsinglelayer neural networks.

当隐藏层被添加到单层神经网络时，从而产生了多层神经网络。

When hidden layers are added to asingle-layer neural network, this produces a multi-layer neural network.

因此，多层神经网络由输入层、隐藏层和输出层组成。

Therefore, the multi-layer neural networkconsists of an input layer, hidden layer(s), and output layer.

具有单个隐藏层的神经网络被称为浅层神经网络或没有新意的神经网络。

The neural network that has a single hiddenlayer is called a shallow neural network or a vanilla neural network.

包含两个或多个隐藏层的多层神经网络被称为深度神经网络。

A multi-layer neural network that containstwo or more hidden layers is called a deep neural network.

在实际应用中使用的大多数神经网络都是深度神经网络。

Most of the contemporary neural networksused in practical applications are deep neural networks.

下表总结了依赖于分层结构的神经网络分支。

The following table summarizes the branchesof the neural network depending on the layer architecture.

将多层神经网络按以上两种类型进行分类的原因与其发展的历史背景有关。

The reason that we classify the multi-layerneural network by these two types has to do with its historical background ofdevelopment.

神经网络从单层神经网络发展到浅层神经网络，再到深层神经网络。

The neural network started as thesingle-layer neural network and evolved to the shallow neural network, followedby the deep neural network.

自从浅层神经网络出现二十年后，直到二十世纪中叶深层神经网络才受到重视。

The deep neural network has not beenseriously highlighted until the mid-2000s, after two decades had passed sincethe development of the shallow neural network.

因此，长期以来，多层神经网络仅仅意味着单隐层神经网络。

Therefore, for a long time, the multi-layerneural network meant just the single hidden-layer neural network.

——本文译自Phil Kim所著的《Matlab Deep Learning》

更多精彩文章请关注微信号：