Linear and Nonlinear Judgment in Machine Learning
Speaking of linearity and nonlinearity, is your intuitive understanding like this:
But this intuitive understanding cannot actually answer the following question:
So why is the convolution operation linear and ReLU non-linear?
Many people are not very clear about the definition of linearity.
In fact, the definition of linearity is: F(ax+y) = aF(x) + F(y)
,
where x, y are variables and a is a constant.
The convolution operation satisfies the above formula, so it is a linear operation.
The following can quickly determine the three common cases of nonlinearity:
(变量)^n
, and n is not 1|变量|
A variable with an absolute value is nonlinearsgn(变量)
have variables inside symbolic functions
And the ReLU looks like this:
So, it is clear that ReLU is nonlinear.