【读书1】【2017】MATLAB与深度学习——正视过度拟合(2)

交叉验证可以保持验证数据集的随机性，可以更好地检测模型的过度拟合。

As the cross-validation maintains therandomness of the validation dataset, it can better detect the overfitting ofthe model.

图1-11描述了交叉验证的概念。

Figure 1-11 describes the concept ofcross-validation.

图1-11 交叉验证示意图Cross-validation

图中的黑色阴影部分表示验证数据集，在整个训练过程中进行随机选择。

The dark shades indicate the validationdata, which is randomly selected throughout the training process.

机器学习的类型（Types of machine learning）

目前已经开发了许多不同类型的机器学习技术来解决各种领域中的问题。

Many different types of Machine Learningtechniques have been developed to solve problems in various fields.

根据训练方法的不同，这些机器学习技术可分为三种类型（见图1-12）。

These Machine Learning techniques can beclassified into three types depending on the training method (see Figure 1-12).

图1-12 三种类型的机器学习技术Three types of MachineLearning techniques

第一种：有监督的学习

Supervised learning

第二种：无监督的学习

Unsupervised learning

第三种：强化学习

Reinforcement learning

有监督学习与人类学习事物的过程非常相似。

Supervised learning is very similar to theprocess in which a human learns things.

考虑人类在解答习题时获取新知识的情况。

Consider that humans obtain new knowledgeas we solve exercise problems.

1. 选择一道练习题。

2. Select anexercise problem.

运用现有知识解决该问题。

Apply current knowledge to solve theproblem.

将自己的答案与解决方案相比较。

Compare the answer with the solution.

1. 如果自己的答案是错误的，修正自己当前的知识。

2. If theanswer is wrong, modify current knowledge.

3. 对所有练习题重复第1步和第2步

4. Repeat Steps 1 and 2 for all theexercise problems.

当将此示例与机器学习过程进行类比时，练习题和解决方案对应于训练数据，而知识对应于模型。

When we apply an analogy between thisexample and the Machine Learning process, the exercise problems and solutionscorrespond to the training data, and the knowledge corresponds to the model.

重要的是我们需要标准解决方案。

The important thing is the fact that weneed the solutions.

这是有监督学习的重要方面。

This is the vital aspect of the supervised learning.

“有监督学习”的名字甚至暗示了老师给学生解答习题的辅导。

Its name even implies the tutoring in whichthe teacher gives solutions to the students to memorize.

在有监督学习中，每个训练数据集应该包括输入和正确的输出组成的输入输出对。

In supervised learning, each trainingdataset should consist of input and correct output pairs.

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

更多精彩文章请关注微信号：