Many students have a headache when they see math. In fact, the basic mathematical principles of the model are not difficult, and the difficult formulas can be solved one by one when they are encountered in the later models. Remember not to go deep into math knowledge in the early stage!

(1) high number

First of all, don't be afraid of math. When we encounter a mathematical formula that we don't know, we should think more about what problems this formula can solve, rather than obsessing about the derivation and calculation of the formula.

For example, the fundamental principle of gradient descent and back propagation is derivative, and the global optimal solution is the extreme point, so the optimal solution must be at some inflection point of derivative, and I learned a lot of similar knowledge in high school.

(2) Line generation

In deep learning, the most important application of line generation is the multiplication of high-dimensional data, which can greatly improve the operation speed.

First of all, we should understand the meaning of each dimension of the matrix.

Secondly, the operation rules of the matrix are clear.

(3) Probability theory

The knowledge of probability theory seems less important in AI system, but it is everywhere. From data preprocessing, modeling, model parameter initialization and normalization to final result analysis, it is closely related to probability theory.

Many commonly used indicators are familiar to everyone, such as mean and variance. If you don't have a strong knowledge of probability theory before, I suggest that after you master common sense, you should get into the habit of asking questions when you encounter problems you don't understand in practice. This part of knowledge is not too difficult, but it is very important to understand the model and process:

For example, why do tree models generally do not need data normalization?

Is data normalization necessary in logistic regression, neural network and PCA?