· How does Xiaobai quickly start deep learning

Personal Home –> xiaosongshine. Github. IO /

With the rise of deep learning in recent years, many researchers have been involved in this field. As universities have put their courses online, many learning resources and online courses have emerged, and many large companies, such as Google and Facebook, have put their open source frameworks on Github. It makes deep learning more and more simple. Many troublesome repetitive operations have been simplified through the framework, which also allows everyone to have access to deep learning.

There are all kinds of learning experience sharing online, some people pay attention to the accumulation of theoretical knowledge, read a lot of books, but the hands-on experience is zero; There are also some people who are passionate about code implementation and study code written by others every day. For both cases, I think that is bad, the depth of the study is the combination of theory and engineering fields, not only need to write code capability is strong, also needs to have the theoretical knowledge to read papers, realize the thesis put forward a new idea, so that the course of our study should be theory combined with the code, the learning tasks on both sides of the balance, Can not appear only – side and not learn the other – side of the situation, because only theory and code to avoid learning in-depth, you will find that they will have a lot of knowledge loopholes.

Before you can get into the field of deep learning, you need to have a basic knowledge of Python, calculus and linear algebra. Here are some suggestions for learning deep learning.

Calculus and linear algebra don’t require much knowledge. For integrals you just know the derivative and the partial derivative, for linear derivatives you just know the matrix multiplication.

Programming based

For Python, there are three resources to start deep learning programming after learning the first resource. The next two resources will help you understand Python and its numerical calculations in greater depth.

(1) The Stupid Method Python

This book is aimed at zero-based scholars, through a series of simple examples’ quick people door basic Python operations.

This series of tutorials can be used to learn Python in a more comprehensive way, just by mastering the basics of Python in the first few chapters. The later parts are more specialized Web development content, which is not required for machine learning.

(3) Edx: JntroductIon (0 Computer Science and Programming Using Python

This is MIT’s public spy, a concise and comprehensive presentation of computer science in Python, suitable for further study.

Theoretical basis

(1) Linear algebra

Linear algebra is equivalent to the cornerstone of deep learning, which has a large amount of short matrix operations, and some ideas of matrix decomposition of linear algebra have been used for reference in machine learning, so it is necessary to master linear algebra. Can refer to the following information tour, learning:

• This is how Linear Algebra should Be learned

• MIT linear Algebra Open course

• Coding The M alpha trix

(2) Fundamentals of machine learning

Although deep learning is very popular now, it also needs to master its root, namely machine learning, which is the essence and core. The learning resources here are ranked from easy to difficult:

• AndrewNg’s introduction to machine learning on Coursera

• The foundation of machine learning and machine learning techniques of Lin Xuen-guo

• Udacity’s nanodegree in machine learning

• Machine Learning by Zhou Zhihua

• Statistical Learning Methods by Li Hang

• Pattern Recognition and Machine Learning

(3) Deep learning

This is the most active research field in recent years, with many revolutionary breakthroughs and cutting-edge learning resources, such as:

• Two deep learning courses at Udacity

• Coursera Neural input {etworksfor Machine Learning

• cs231n Stanford

• cs224n Stanford