How to teach yourself artificial intelligence

The general steps for learning AI:

(1) Understand some background knowledge of artificial intelligence;

(2) Supplement mathematics or programming knowledge;

(3) Be familiar with the machine learning tool library;

(4) Systematically learn AI knowledge;

(5) Do some AI applications by yourself;

1 Understand the background knowledge of artificial intelligence

There are many concepts in artificial intelligence, such as machine learning, deep learning, neural networks, etc., which make beginners feel that artificial intelligence is mysterious and difficult to understand. When you first start learning, you just need to know the general meaning of these terms. You don’t need to delve too deeply into it. After studying for a period of time, you will naturally understand what these concepts specifically represent.

Artificial intelligence is an interdisciplinary subject, among which mathematics and computer programming are the two most important aspects of learning artificial intelligence. These articles "Understanding Artificial Intelligence" before the "Zhiyun AI Column" have also been introduced to you. Students who have not read them can read them.

The following figure shows the general route of artificial intelligence learning:

2 Supplementary mathematics or programming knowledge

For engineers who have graduated, before systematically learning AI , generally need to add some knowledge of mathematics or programming. If you are good at math and programming, learning artificial intelligence will be much easier.

Many students are afraid when it comes to mathematics. However, it can be said that mathematics cannot be bypassed when learning artificial intelligence. At the entry level, you don’t need too advanced mathematics, mainly advanced mathematics, linear algebra and probability theory. In other words, the mathematics knowledge learned in freshman and sophomore year is completely sufficient. If you want to work as a machine learning engineer or engage in artificial intelligence research, you should learn more mathematics. Being good at mathematics will be a major advantage at work.

Python is very popular in the field of machine learning. It can be said to be the most used programming language, so Python programming also needs to be mastered. Among many programming languages, Python is relatively easy to learn and use. Learning Python well will also benefit a lot.

3 Be familiar with machine learning tool libraries

Nowadays, people implement artificial intelligence, mainly based on some machine learning tool libraries, such as TensorFlow, PyTorch, etc.

I recommend everyone to learn PyTorch here. PyTorch is very popular and is an easy-to-use machine learning tool library. Some people commented that PyTorch "can't say how good it is, but it is very comfortable to use."

When you first start learning artificial intelligence, you can first run the examples on the official website of the tool library, such as MNIST handwriting recognition, etc. In this way, you will have a perceptual understanding of artificial intelligence and eliminate the initial strangeness. Then you can look at the code inside, and you will find that the program of the neural network is not complicated, but you will have many questions about the principles and training of the neural network. This is a good thing, because studying with questions is more productive.

4 Systematic Learning Artificial Intelligence

The artificial intelligence here mainly refers to machine learning, because currently artificial intelligence is mainly achieved through machine learning.

There are three main parts of machine learning knowledge:

(1) Traditional machine learning algorithms, such as decision trees, random forests, SVM, etc., these are called traditional machine learning algorithms and are Relative to deep learning.

(2) Deep learning refers to deep neural networks, which can be said to be the most important and core artificial intelligence knowledge at present.

(3) Reinforcement learning, derived from cybernetics, is sometimes translated as reinforcement learning. Deep learning can be combined with reinforcement learning to form deep reinforcement learning.

What you need to know here is that deep learning is not difficult to learn. For some engineering graduate students, it usually only takes a few weeks to get started and train some neural networks for practical applications. But it is not easy to have a deep understanding of in-depth learning, and it usually takes several months.

There are many types of traditional machine learning algorithms, and some algorithms have a lot of mathematical formulas, such as SVM. These algorithms are not easy to learn, so you can learn deep learning first, and then slowly supplement these traditional algorithms.

Reinforcement learning is more difficult. It usually takes two or three months of continuous learning to gain some understanding.

5 Start making some AI applications

After learning deep learning for a few weeks, you can start trying to make some AI applications, such as image recognition, style transfer, and text poetry. Generate etc. The effect of learning while practicing will be much better, and your understanding of neural networks will gradually deepen.