正确的输出是模型对于给定的输入应该产生的结果。

The correct output is what the model issupposed to produce for the given input.

{输入，正确的输出}

{ input, correct output }

有监督学习中的学习是对模型进行一系列的修改，以减少同一输入下正确输出与模型输出之间的差异。

Learning in supervised learning is theseries of revisions of a model to reduce the difference between the correctoutput and the output from the model for the same input.

如果模型被完全训练，它将产生与训练数据的输入相对应的正确输出。

If a model is perfectly trained, it willproduce a correct output that corresponds to the input from the training data.

与此相反，无监督学习的训练数据只包含没有正确输出的输入。

In contrast, the training data of theunsupervised learning contains only inputs without correct outputs.

{输入}

{ input }

乍一看，似乎很难理解，没有正确的输出，该如何进行训练呢。

At a first glance, it may seem difficult tounderstand how to train without correct outputs.

然而，目前已经开发了许多这种类型的无监督学习方法。

However, many methods of this type havebeen developed already.

无监督学习通常用于研究数据的特征和数据的预处理。

Unsupervised learning is generally used forinvestigating the characteristics of the data and preprocessing the data.

无监督学习概念类似于一个学生，他仅仅通过构造和属性来分类问题，而没有学习如何解决这些问题，因为没有已知的正确输出。

This concept is similar to a student whojust sorts out the problems by construction and attribute and doesn’t learn howto solve them because there are no known correct outputs.

强化学习采用输入、输出、评分等集合作为训练数据。

Reinforcement learning employs sets ofinput, some output, and grade as training data.

强化学习通常在需要最佳交互时使用，例如控制和游戏。

It is generally used when optimalinteraction is required, such as control and game plays.

{输入，一些输出，输出评分}

{ input, some output, grade for this output}

本书只涉及监督学习。

This book only covers supervised learning.

与无监督学习和强化学习相比，监督学习的应用更为广泛，更重要的是，它是你在进入机器学习和深度学习的世界时需要学习的第一个概念。

It is used for more applications comparedto unsupervised learning and reinforcement learning, and more importantly, itis the first concept you will study when entering the world of Machine Learningand Deep Learning.

分类与回归（Classification and Regression）

有监督学习的两种最常用类型是分类和回归。

The two most common types of application ofsupervised learning are classification and regression.

这些术语听起来可能不熟悉，但实际上并不那么具有挑战性。

These words may sound unfamiliar, but areactually not so challenging.

我们从分类开始谈起。

Let’s start with classification.

这可能是机器学习中最普遍的应用。

This may be the most prevailing applicationof Machine Learning.

从字面上来说，分类问题的重点是找到数据所属的类别。

The classification problem focuses onliterally finding the classes to which the data belongs.

举例说明可能会有所帮助。

Some examples may help.

垃圾邮件过滤服务➔将邮件分类为普通邮件或垃圾邮件

Spam mail filtering service ➔ Classifies the mails by regular or spam

数字识别服务➔将数字图像分类为0—9

Digit recognition service ➔ Classifies the digit image into one of 0-9

人脸识别服务➔将人脸图像分类为注册用户之一

Face recognition service ➔ Classifies the face image into one of the registered users

我们在上一节中谈到，有监督学习需要用于训练数据的由输入和正确输出组成的输入输出对。

We addressed in the previous section thatsupervised learning requires input and correct output pairs for the trainingdata.

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

更多精彩文章请关注微信号：