Higher degree polynomial

[polodealvarado]

I have a couple of questions about your paper "Polynomial Regression as an Alternative to Neural Networks".

• At point 5 it is said that "many authors recommend not using polynomial models of degree higher than 2 or 3" .

In which cases is it possible to get higher degree polynomials which fit well?

• At point 6.1 it is said that " the degree of the approximating polynomial increases from layer to layer".

So, if I get a polynomial of degree 3 which fits well, then the corresponding neural network will have three layers approximately?