Why does learning artificial intelligence pay so much attention to mathematical foundation?

This must start with the essence of artificial intelligence at present. At present, the deep learning system based on neural network can actually be regarded as a linear algebraic matrix model and a differential equation from a microscopic point of view.

The focus of artificial intelligence is intelligence, and the ultimate embodiment of intelligence should be randomness. For example, you never know what an independent intelligent life will do next second.

Mathematics has a solution, which can be calculated, and intelligence has no solution, which is unpredictable, but many behaviors of intelligence can be calculated by mathematics, so there should be a strong relationship between intelligence and mathematics, but it is not the only correlation.

This is why most research institutes at home and abroad pay attention to mathematical ability first when recruiting interns.

What kind of mathematical foundation is needed to learn artificial intelligence?

Among the three mathematics courses of Linear Algebra, Probability Theory and Optimization Theory, the first two courses are modeling and the last one is solving, which is the basis of learning artificial intelligence. I have everything you want.

1. linear algebra

Linear algebra is an essential knowledge in the process of learning artificial intelligence. We are most familiar with simultaneous equations in linear algebra, and the origin of linear algebra is to solve simultaneous equations. Only with the deepening of research, people find that it has a wider range of uses.

2. Probability theory

Probability and Statistics is an important basic course in statistical learning, because machine learning is often dealing with the uncertainty of transactions.

optimize

After the model is established, how to solve this model belongs to the category of optimization. Optimization is to find an optimal solution when the analytical solution of the problem cannot be obtained. Of course, it is necessary to define what is the best in advance, just like defining the rules of the game before a basketball game.

The usual practice is to find a way to construct a loss function, and then find the minimum value of the loss function to solve.