Finished Translation ai-edu/A-基础教程/A2-神经网络基本原理/en-us/Step1 - BasicKnowledge/00 Preface.md (#644)

* Update README.md

* Update 03.1-Mean Square Error(MSE) Loss Function.md

* Update 00 前言.md

* Update 00 前言.md

* Update 00 前言.md

* Update 00 前言.md

* Update 00 前言.md

* Update 00 前言.md

* Create 00 Preface.md

* Update 00 Preface.md

* Rename 00 Preface.md to 000 前言.md

* Rename 00 前言.md to 00 Preface.md

* Update 00 Preface.md

* Update 00 Preface.md

* Rename 000 前言.md to 00 前言.md

* Create 14 Step Seven - Deep Neural Network.md

A-基础教程/A2-神经网络基本原理/en-us/Step1 - BasicKnowledge/00 Preface.md

@@ -0,0 +1,94 @@
+<!--Copyright © Microsoft Corporation. All rights reserved.
+  适用于[License](https://github.com/Microsoft/ai-edu/blob/master/LICENSE.md)版权许可-->
+
+# Preface
+
+**"What I cannot create, I do not understand." (American physicist Richard Feynman)**
+
+In today's technologically advanced world, a lot of knowledge can be found on the Internet. However, the quality of blogs/articles is questionable, its focus is not clear, or it is directly copied from other people’s blogs. This situation makes it difficult for most beginners to learn. They may even be led astray, increasing the steepness of the learning curve. Of course, there are creators who are very responsible. The quality of their articles may be very high, but its scope is not enough. When you are enjoying yourself, there are no follow-up chapters and so it cannot be a good learning system.
+
+Beginners can choose to read textbooks or books on theory, but the chicken and the egg problem arises: If you don’t understand, then after reading you still won’t understand the theory; if you understand, then you don’t need to read the theory. This is a shortcoming of some textbooks and theoretical books.
+
+The author has seen professor Andrew Ng's (Wu Enda) classes. The theoretical knowledge is simple yet deep, and is very clear. Although there are very few code examples, I still strongly recommend looking into Andrew Ng's classes. The author's experience is that Andrew Ng's videos can be saved on a mobile device, and you can watch and learn on your own time.
+
+There are external reasources that you can use to study deep learning online. The author took part in several group purchases. The teachers and teaching assistants are generally very responsible. Lastly, you can rewatch the videos and download the powerpoint course materials. These courses mainly focus on engineering projects, and explain the use of deep learning frameworks and tools. That is, teaching you how to use helpful tools for modeling, training, etc. But for beginners, understanding a new concept may require a lot of previously attained knowledge. An extremely steep learning curve can be frustrating. Maybe someone knows X but doesn't know why. They could end up as an assistant engineer, and their career development ends up being restricted.
+
+There is a proverb: Teach him how to fish and you feed him for his life time. After experiencing the mentioned learning experience, the author, who is a programmer, eagerly feels that a learning experience should be done in "learning by doing". By reproducing some basic theories in writing code, you can deeply understand its meaning. Then you can expand and extend so readers have the ability to draw inferences.
+
+The author, who has summarized their learning experience has also summarized the introductory knowledge of deep learning into 9 steps. These steps are referred to as the 9-step learning method:
+
+1. Basic concepts
+2. Linear regression
+3. Linear classification
+4. Nonlinear regression
+5. Non-linear classification
+6. Model inference and deployment
+7. Deep neural networks
+8. Convolutional neural networks
+9. Recurrent neural networks
+
+Many books I have seen start directly from Step 7. Their assumption is that the reader has mastered previous knowledge. However, for beginners starting from scratch, this assumption is not correct.
+
+In the following explanations, we will generally use the following methods:
+
+1. Ask the question: First ask a hypothetical question related to reality. In order to move from simple stuff to deeper stuff, these questions should not complicated. They are simplified versions of actual engineering problems.
+2. Solution: Use the knowledge of neural networks to solve these problems. Start from simple and basic models and build step by step to achieve a complex model.
+3. Principle analysis: Use basic physics concepts or mathematical tools to understand how neural networks work.
+4. Visual understanding: Visualization is an important aspect of learning new knowledge. As we use simple examples, they can easily be visualized.
+
+Principle analysis and visual understanding are also features of this book - trying to make neural networks interpretable instead of blindly using them.
+
+This is also a very important point. We have supporting Python code. In addition to some necessary scientific computing libraries and plotting libraries, such as NumPy and Matplotlib, we did not use any existing deep learning frameworks, but led everyone from scratch. Start to build your own simple knowledge system to understand the many knowledge points in deep learning, and slowly scale to more complex ones.
+
+For our friends who have no experience in Python, reading the sample code can also help you learn Python, which can kill two birds with one stone. As the difficulty of a problem deepens, the code will grow as well. The before and after have a connection much like inheritance. The final code will form a small framework, which I call a Mini-Framework. They can be created by calling building block functions to build deep learning components.
+
+The code has been written and debugged by the author himself. Each chapter can run independently, and the results described in the relevant chapters can be obtained, including printouts and graphical output.
+
+In addition, for ease of understanding, the author draws a large number of schematic diagrams. The number totals to more than 10 times the amount of similar books. A picture is worth a thousand words. I believe that everyone will quickly and completely understand the knowledge points I want to share through these diagrams. This way, everyone can start from the real "zero", have a basic understanding of neural networks and deep learning, and be able to practice.
+
+Requirements for readers:
+
+1. Have learned linear algebra and differentiation in advanced mathematics
+2. Have programming fundamentals. You don't need to know Python because you can learn it from the sample code
+3. Thinking + hands-on learning mentality
+
+To help the reader achieve a level of greatness:
+
+1. Judge which tasks are achievable by machine learning and which are science fiction. Don’t stay an amateur
+2. Thoroughly understand the basic theories of neural networks and deep learning
+3. Cultivate the ability to solve practical problems by analogies
+4. Have the ability to learn more complex models and advanced content alone
+5. For the more experienced readers, they can cultivate the ability to develop new models
+
+## Notation and Convention
+
+|Symbol|Meaning|
+|---|---|
+|$x$|Sample training value|
+|$x_1$|The first sample/characteristic value of the sample which will be explained under the context|
+|$x_{12},x_{1,2}$|The second eigenvalue of the first sample|
+|$X$|Multi-sample training matrix|
+|$y$|Sample label value for training|
+|$y_1$|Label value of the first sample|
+|$Y$|Multi-sample label matrix for training|
+|$z$|Linear operation's resulting value|
+|$Z$|Linear operation's resulting matrix|
+|$Z1$|Linear operation's resulting matrix of the first layer network|
+|$\sigma$|Activation function|
+|$a$|Activation function's resulting value|
+|$A$|Activation function's resulting matrix|
+|$A1$|Activation function's resulting matrix of the first layer network|
+|$w$|Parameter weight value|
+|$w_{12},w_{1,2}$|The weight value at the first row and second column in the parameter weight matrix|
+|$w1_{12},w1_{1,2}$|The weight value of the first row and second column in the parameter weight matrix of the first layer network|
+|$W$|Parameter weight matrix|
+|$W1$|The weight parameter matrix of the first layer network|
+|$b$|Offset parameter value|
+|$b_1$|The irst offset value in the parameter offset matrix|
+|$b2_1$|The first offset value in the parameter offset matrix of the second layer network|
+|$B$|Offset parameter matrix (vector)|
+|$B1$|Offset parameter matrix of the first layer network (vector)|
+|$X^T$|X's transpose matrix|
+|$X^{-1}$|X's inverse matrix|
+|$loss,loss(w,b)$|One-sample error function|
+|$J, J(w,b)$|Multi-sample loss function|

A-基础教程/A2-神经网络基本原理/en-us/Step1 - BasicKnowledge/00 前言.md

@@ -1,21 +1,23 @@
 <!--Copyright © Microsoft Corporation. All rights reserved.
   适用于[License](https://github.com/Microsoft/ai-edu/blob/master/LICENSE.md)版权许可-->
 
-# Preface
+# 前言
 
-**"What I cannot create, I do not understand." (American physicist Richard Feynman)**
+**What I cannot create, I do not understand.**
 
-In today's technologically advanced world, a lot of knowledge can be found on the Internet. However, the quality of blogs/articles is questionable, its focus is not clear, or it is directly copied from other people’s blogs. This situation makes it difficult for most beginners to learn. They may even be led astray, increasing the steepness of the learning curve. Of course, there are creators who are very responsible. The quality of their articles may be very high, but its scope is not enough. When you are enjoying yourself, there are no follow-up chapters and so it cannot be a good learning system.
+**如果我不能亲手搭建起来一个东西，那么我就不能理解它。 -- 美国物理学家理查德·费曼**
 
-Beginners can choose to read textbooks or books on theory, but the chicken and the egg problem arises: If you don’t understand, then after reading you still won’t understand the theory; if you understand, then you don’t need to read the theory. This is a shortcoming of some textbooks and theoretical books.
+在互联网发达的今天，很多知识都可以从网络上找到，但是网络上的博客、文章的质量参差不齐，或者重点不明确，或者直接把别人的博客抄袭过来。这种状况使得广大的初学者们学习起来很困难，甚至误入歧途，增加了学习曲线的陡峭程度。当然也有很多博主非常非常负责任，文章质量很高，只是连续度不够，正看得过瘾的时候，没有后续章节了，无法形成知识体系。
 
-The author has seen professor Andrew Ng's (Wu Enda) classes. The theoretical knowledge is simple yet deep, and is very clear. Although there are very few code examples, I still strongly recommend looking into Andrew Ng's classes. The author's experience is that Andrew Ng's videos can be saved on a mobile device, and you can watch and learn on your own time.
+初学者也可以选择看一些教材或者理论书籍，但是，一个鸡生蛋蛋生鸡的问题出现了：如果你不懂，那么看完了理论你还是不会懂；如果你懂了，那么你就没必要看理论。这也是很多教材或者理论书籍的缺憾。
 
-There are external reasources that you can use to study deep learning online. The author took part in several group purchases. The teachers and teaching assistants are generally very responsible. Lastly, you can rewatch the videos and download the powerpoint course materials. These courses mainly focus on engineering projects, and explain the use of deep learning frameworks and tools. That is, teaching you how to use helpful tools for modeling, training, etc. But for beginners, understanding a new concept may require a lot of previously attained knowledge. An extremely steep learning curve can be frustrating. Maybe someone knows X but doesn't know why. They could end up as an assistant engineer, and their career development ends up being restricted.
+笔者也看过吴恩达老师的课，理论知识讲得由浅入深，还是非常清楚的，虽然代码示例基本没有，但仍然强烈建议大家去看。笔者的心得是：视频可以事先缓存在手机中，利用一些时间片段就可以学习了。
 
-There is a proverb: Teach him how to fish and you feed him for his life time. After experiencing the mentioned learning experience, the author, who is a programmer, eagerly feels that a learning experience should be done in "learning by doing". By reproducing some basic theories in writing code, you can deeply understand its meaning. Then you can expand and extend so readers have the ability to draw inferences.
+社会上还有一些网课，在线讲解深度学习的知识，笔者也参加了几个团购，老师和助教一般都很负责任，最后可以回看录像，下载PPT课件。这些课程一般偏重于工程项目，讲解深度学习框架和工具的使用，即教大家如何使用工具建模、训练等等，也是很有帮助的。但对于初学者来说，理解一个新概念可能需要前面很多个已有知识点的支撑，门槛过高，一下子就变得很沮丧。或者是知其然而不知其所以然，最后沦为调参工程师，职业发展受到了限制。
 
-The author, who has summarized their learning experience has also summarized the introductory knowledge of deep learning into 9 steps. These steps are referred to as the 9-step learning method:
+还是应了那句古话：授人以鱼不如授人以渔。经历了以上那些学习经历，程序员出身的笔者迫切感觉到应该有一种新的学习体验，在“做中学”，用写代码的方式把一些基础的理论复现一遍，可以深刻理解其内涵，并能扩充其外延，使读者得到举一反三的泛化能力。
+
+笔者总结了自身的学习经历后，把深度学习的入门知识归纳成了9个步骤，简称为9步学习法：
 
 1. 基本概念
 2. 线性回归

A-基础教程/A2-神经网络基本原理/en-us/Step7 - DNN/14 Step Seven - Deep Neural Network.md

@@ -0,0 +1,12 @@
+<!--Copyright © Microsoft Corporation. All rights reserved.
+  适用于[License](https://github.com/Microsoft/ai-edu/blob/master/LICENSE.md)版权许可-->
+
+# Step 7: Deep Neural Networks
+
+## Summary
+
+In previous steps, we used simple cases to gradually learn a lot of knowledge, making it easier for us to touch upon deep learning.
+
+Starting from this step, we will discuss some details of deep learning. This includes weight matrix initialization, gradient descent optimization algorithms, batch normalization, and other advanced concepts.
+
+Due to the strong learning ability of the deep network, it will cause the network to overfit the sample data, resulting in insufficient generalization abilities. Therefore, we need some means to improve the generalization ability of the network.