Regularization is also called regularization, also known as penalty term.
Role: Regularization is to avoid overfitting.
Principle: According to the Taylor expansion in high school mathematics, any function can be approximated by a polynomial. The process of solving the parameters of the polynomial is the process of model training in machine learning. The model trained in machine learning is essentially the parameters of the polynomial. The solution process, how to solve the parameters of the polynomial so that the solved polynomial is more robust is a good model.
Overfitting is the essence of poor robustness
Avoid over-fitting: the most intuitive is to reduce the number of polynomial parameters, or the minimum number of polynomial parameters is the best effect, the first model is small, and the second model has few parameters, it will not appear. Fitting, robustness is good.
How to solve the minimum number of parameterized items w, that is, the maximum 0 norm
0 norm: the number of non-zero elements in the vector
https://www.zhihu.com/question/20924039